Elad Blog

AI Market Clarity

Elad Gil — Tue, 22 Jul 2025 16:47:51 GMT

AI Markets Have Crystalized

AI markets have evolved significantly over the last 4 years. When GPT-3 came out and scaling laws were openly discussed in the AI literature, it seemed clear that you could extrapolate the rate of progress from GPT-2 to GPT-3 onwards through GPT-4, 5 etc and realize a revolution was going to happen.

4 years ago, I started looking for Generative AI companies to back or help start given this curve. I ended up leading or participating in early rounds in companies like Harvey, Perplexity, Character.AI, BrainTrust, and others. At the time it was clear to just “back all the best people working on all the biggest problems” because very few people were actually starting generative AI companies. OpenAI seemed to be the only clear foundation model company (with Anthropic pre-launch and promising, Llama non-existent, and Google clearly someone who could (and would eventually) aggressively innovate but was stymied at the time by its own internal processes).

As more people outside of the core AI community woke up to this opportunity, or researchers and engineers from the main labs left to start new companies in AI, the world of AI became murkier. I used to say that the more I learn about AI, the less I knew about AI – because it was unclear in many early markets who the likely winners would be and the underlying models and tech were changing so fast. For example in 2022, it was clear code / AI driven software engineering was going to be important, but it was unclear who the winners would be (for example Cursor was not launched until 2023, Codium launched Windsurf in GA about 9 months ago, and Cognition launched Devin in limited release a bit over a year ago).

We have now entered an era where the first set of AI markets have solidified and a likely set of winners have emerged. This does not mean others won’t show up over time to compete in these markets or that current leaders won’t get acquired or eventually die (just as Stripe launched over a decade after PayPal and 4 or so years after Braintree, and Facebook launched a few years after Friendster and Myspace). We will also see a new set of markets crystalize in the coming few years and these markets today seem quite uncertain.

Markets with more clarity

1. Foundation Models- LLMs

There are many types of foundation models including large language models (LLMs), as well as models for voice, images, video, music, chemistry, biology, materials, physics and other areas. Foundation models are often driven by scale (of data, compute, certain types of post training and feedback, etc). Scale means capital, so to win in the LLM market you need high availability of capital now entering the many billions.

In the LLM market, a core set of companies have clearly emerged as the ongoing players of the future. They are often partnered with hyperscalers (Amazon with Anthropic, Google GCP with Gemini, Microsoft Azure with OpenAI and its own efforts) as these companies have an economic incentive (cloud spend on AI via AI adoption) to fund these companies that is independent of whether these companies are good investments or not (they often are). Revenue ramps for foundation model companies are rumored to be in the $0 to many $billions range in just 3 or so years, while cloud spend on “AI” has been reported to have reached a few billion per quarter for some of the main clouds.

The core players in the LLM world are now Anthropic, Google, Meta (via Llama), Microsoft, Mistral, OpenAI, X.AI. Three or four of these companies are the clear winners on various benchmarks and most broadly adopted by developers and enterprise, and are also driving most of the spend in the industry. There are newer entrants like SSI and Thinking Machine Labs driven by brilliant AI researchers, which may either come up with innovative approaches or raise ongoing money to compete, or end up in a few years as acquisitions for companies wishing to enter the market or double down on talent.

In parallel, Chinese companies have launched new open source efforts such as Deepseek, Alibaba Qwen, and more recently Kimi, that perform well on benchmarks. There is a lot to say on Chinese open source LLMs, that might be best covered in a separate post.

In the future, it is unlikely that many new core LLM companies will get started due to capital moats, barring some new breakthrough that somehow doesn’t spread quickly.

Other foundation model markets are still lacking the clear winner or winners although promising companies exist in multiple segments.

2. Code

Code is one of the earliest and clearest large-scale applications of generative AI and LLMs. For example, Github copilot was launched in October 2021 and was already well adopted and useful despite more limited functionality and fidelity of its day. Code has all sorts of properties that make it especially amendable to generative AI approaches. Revenue ramps in code are rumored to be in the $0 to $50 up to $500M in the first 2 years of some of the players live product lifetime – an insanely fast pace.

Like foundation model companies, the core likely winners in code are evident, with a handful of companies seeming to be interesting over time. Large incumbents may still enter the coding market, and products to date do not seem especially sticky but moats tend to come with time. However, this handful of companies will play an important role in the next few years of code barring the unexpected and includes Anthropic’s Claude Code, Cognition / Windsurf, Cursor, Google / Windsurf, Microsoft/Github, OpenAI, potentially startups like Magic or Poolside, and then vibe coding companies like Lovable, Replit, and others. Intriguingly, both Figma and Canva have launched vibe coding tools and one can imagine lots more coming.

A number of questions continue to exist for code (and other markets) including agentic vs IDE-based work flows and their eventual overlap (they will undoubtedly converge), as well as the degree to which the foundation model companies will integrate or subsume more of the coding companies functionality directly given both large economic value, as well as code being a bootstrap path to AGI / SI. These core questions will drive who the final winners in code will be.

3. Legal

The core players in core legal markets have solidified with Harvey (both law firms and enterprise) and CaseText as the current leaders. Other startups working in either overlapping (Legora) or new areas (Crosby) are starting to emerge. EvenUp moved in the post AI era into personal injury and related areas, while Eve and Supio are focused on plaintiff work flows. We are still very early in full workflow automation but startups like Harvey and EvenUp have started to take core legal workflows and build systems to complete the work start from finish. With the centrality of legal, companies like Harvey may be able to naturally integrate into other professional services workflows over time

Given the breadth of legal across vertical areas (patents, contracts, etc.) and type (law firms, enterprise, SMB, consumer) there may be new legal areas yet to explore.

4. Medical Scribing

Physician tools and scribing is another area with a clear set of main players that markets have consolidated against include Abridge, Ambience, Commure / Athelas, and Nuance (Microsoft). Some international players have also emerged in this area and may either end up as stand alone, or consolidated into the main players.

The key next step for these players is to expand products into other areas of the healthcare stack.

5. Customer service / experience

The customer experience market in the US appears to have consolidated in the short term into a few core startup players – Decagon and Sierra, while incumbents like Intercom, Zendesk and others work to add and cross-sell generative AI capabilities. Other startups of interest include Forethought, Maven, Parahelp, Wonderful, and others. Like many of the markets above, customer experience has the characteristic that generative AI is displacing or augmenting humans strongly with agentic work, versus another seat-based workflow tool.

We are starting to see the shift from selling seats, to selling units of cognition. This is an underdiscussed aspect of AI companies in general.

Reasoning model advancements and agentic infrastructure products will only accelerate this shift.

6. Search and IR re-invention

Players focused here include Google, OpenAI (chatGPT), Perplexity, and Meta. Perplexity is notably the main startup in this market, while most of the other players are incumbents. This may end up being true for other consumer and prosumer markets as well, although there is a lot of room for new consumer use cases.

Perplexity and the other players are moving quickly into the agentic future (see section below), both through tools like Deep Research, but also more recently via entry into the browser market. For example, Perplexity’s Comet browser already incorporates agentic actions for shopping on the web and other actions.

Future markets that will be important

3 years ago it was clear foundation models / LLMs, code, healthcare, customer service and other areas would be important AI markets, but it was much less clear who would win or be important. Today the market leaders that are likely to play a short term role are clear (although as mentioned often new startups or incumbents launching products can still join later in the game).

The next set of markets that seem highly interesting and tractable to generative AI include, but are not limited to the below. I and my team are highly interested in investing behind companies in these areas:

Accounting. People are both building software here and in some cases doing rollups.
Compliance. Lots of different forms of compliance. Pharma compliance is one example with early players like Blue Note Health.
Financial tools. Tools to help financial analysts or others. There are a number of great teams and early companies working in this area.
Sales tooling and agents. Lots to do here – from AI agents doing SDR work to tools to augment enterprise sales.
Security. An enormous number of potential applications here, particularly in areas of high service utilization or large team complexity. There will separately emerge security companies focused on AI-end points, agents, or foundation model usage that could lead to data leaks or other exploits.
Other markets TBD. Lots to do!
What have I missed? Let me know on X.com or directly.

There are a set of exciting companies in each of these areas, and which of them will pull ahead or win will likely crystalize in the coming months or quarters. In some cases models may not yet be good enough to address these markets well, in others a broader workflow tool or better GTM approach needs to be developed. In some cases it might just be a matter of time – it takes time to fully understand a customer, build against their needs, and see product-market pull and fit.

I (and my team) am actively looking to invest in all these areas – so please ping if working on it. :)

Model vs GTM vs Team?

For the “new” market segments above, one core question is what has been preventing market crystalization?

Part of it for some markets is the models need to advance in reasoning or fidelity. For example, legal workflows did not work on GPT 3.5 but with GPT4 really took off and Harvey benefited from this plus custom model work. Similarly, coding tools like Cursor benefited from Claude 3.5 (launched June 2024) and were just not as useful until the models used crossed some tipping point of fidelity. Building early against customers pre-model fidelity allows you to capture market share once the models get better.

This leads to the concept of the GPT Ladder - at a certain level of GPT (or Claude, or Gemini, or Grok, or Llama) release, specific new markets will open up. For example, GPT-5 (or Claude X, or Gemini Y) should support an entirely new market that simply was not tractable technologically in the absence of the model.

Beyond models, other things that may prevent adoption of generative AI in a market may include:

Go-to-market, buying behavior & competition.
- Startups may be getting go-to-market wrong (selling to the wrong customers in the wrong way)
- There may be too much incumbent lock in (or ability to counter startups quickly with worse product or cross sell against an existing UI or workflow owned by the incumbent).
- Buyers in the market may just be slow to move and adopt and will get there over time.
Teams. Perhaps the right team has not shown up yet, or the founders are new to the market and will take a bit of time to learn and iterate what customers need.
Time to build. It takes time to build and ship useful products. Sometimes activity in a market is just too nascent and in 6-12 months a big enough product footprint will exist for winner(s) to emerge.

Agents everywhere

One big ongoing shift is moving from pure tooling “AI chat” to agentic workflows. Agents are AI software that do actions on your behalf. It is the difference between looking up information about travel to Spain on Google, and having Google spawn an AI agent that goes and books the travel for you and takes actions on your behalf. Coding tools like Devin and customer service tools like Decagon/Sierra seem the earliest B2B adopters of agentic workflows, while informational tools like chatGPT, Gemini, and Perplexity are adding agents to do deep research for their users.

As reasoning models and agents proliferate, new infrastructure to support agentic deployment & workflows is accelerating. A number of startups are working on agentic frameworks or infrastructure, while consulting or large-scale deployment firms are also adding agents to their roster of tools to deploy in enterprises.

We are starting to see the shift from selling seats, to selling units of cognition or human equivalent labor.

AI Roll Ups

I have been talking about, and investing in, generative AI driven roll ups for 3 years or so. From the earliest days of generative AI it was clear that this new form of scaled transformer-based AI was very good at human knowledge work – which is much of the white collar services economy. In AI driven roll ups, buying a company versus just selling them software can lead to outsized faster adoption and economics than just selling software.

Often, adoption of AI is not a technology problem – it is an organization, process, and people issue. Can you rework an entire organization or its way of doing things around the AI tooling? This is often a much harder problem than just the AI tooling itself and requires owning the actual company to be able to rework organizational processes on a short enough time frame to matter (at least for a startup – incumbents often take a long time).

I have funded 2 AI driven buyouts to date and am excited to back more- reach out if you are working in this area!

Market ending moves

In another post, I talk about “market ending moves” – what big strategic play can you do (M&A? Enormous capital scale and investment? Other?) can you do to just win first place in a market.

As markets consolidate, the strategic moves to just win a market become clear. This will likely lead to various forms of M&A, partnership, channel lock in, or other strategies as the handful of players per market consolidate down into one or a small handful.

We should see a lot of consolidation and M&A coming quite soon due to the ability to win a market by combining the two main startup leaders (often hard to negotiate but worth doing) or incumbent/startup pairs (distribution + tech = winning).

Very exciting times are ahead.

Summary - AI markets have crystalized

AI markets are now the clearest they have been in a few years. The leaders in a given segment are clear for many of the earlier generative AI markets like Code and Legal, while new markets are ripe for disruption. Exciting times ahead.

Biotech companies I wish existed

Elad Gil — Tue, 20 May 2025 23:01:19 GMT

The biopharma ecosystem is set up for a limited number of "indication" or disease areas to be worked on. Part of this is industry structure (many biotech firms are incubated by VC firms to flip into biopharma company roadmaps for M&A), part is intellectual snobbery (“cosmetics of aging - yuck! I want to do Alzheimers”) and part of this is regulatory. A lot of important and commercially viable pre-existing biology that can be translated into real world applications is thus ignored. Here are some things I wish there were more startups working on[1,2,3].

Fertility

Given how important fertility is, it is shocking how few companies are working in the area. Two main thrusts of fertility that are interesting:

Allowing any adult to have children with anyone else. There is strong scientific results out of Japan in mice, in which many cell types can be converted into stem cells, and then differentiated into either sperm or egg cells. The human version of this means (roughly) anyone could have kids with anyone else. Gay couples could have kids with each other (one person would have their cells expanded into sperm, the others eggs), as long as they had a surrogate (see for example this paper from Hayashi’s group in Japan about mice produced via this technique using two fathers). Women of any age could have children. It is a huge unlock societally. Some reviews. Popular version. See also.
Expanding usable egg populations in women. Girls are born with 1 to 2 million oocytes (egg cells) and by puberty end up with ~300,000. Only a small number of these ever mature into the eggs that can be harvested for IVF or used for natural pregnancies. Methods to expand and mature egg cells seem dramatically under developed. See for example.

Anti-aging / longevity

There is clear data that aging is a genetically manipulatable and drugable phenotype. Given the size and important of the prize (e.g. adding 100+ years of productive health life to each person) it is a surprisingly empty field with at most a half dozen or so legitimate companies working on it (including eg BioAge and NewLimit - I am an investor in both).

Neurosensory aging. As you age you lose aspects of hearing and site. For example, the muscle holding the lens of the eye weakens in part leading to blurrier vision and the need for reading glasses for people in their 40s. This seems drugable. Ditto loss of hearing in aging adults. This example of reversing aging in mouse age dependent glaucoma models is intruiging (much simpler approaches then this seem like to work).

Cosmetic aging. Cosmetic use of Botox is roughly at $1.6 billion per year. People are literally injecting a bacterial toxin into their skin. Imagine anti-aging drugs that actually rejuvenate aspects of aging? Examples:

Skin aging and wrinkles. There is enormous demand for things that will reverse or stop skin aging, as seen via botox, face lifts, face fillers, and basic cosmetics.
Balding. Hair loss is age dependent. Some treatments like Minoxidal or Propecia can halt or partially reverse it. Why aren’t there drugs that restore hair completely?
Grey hair. As you age you lose melanin cell production in/around hair follicles. This leads to greying hair. A lot of the biology has been worked out. Why are there not more active efforts to reverse this?

Dental

Tooth regrowth. Many species like sharks can regrow teeth indefinitely, while humans naturally produce a second set of teeth during childhood. Why cant we just regrow a tooth with a bad cavity versus do extensive dental work? Genes like USAG-1 may allow for tooth regrowth in certain animal models, and other approaches and factors exist.

Biomarkers

Often the best way to run a clinical trial or discover drugs rapidly is to have an easily interrogatable biomarker that is a proxy for a biological effect. For example, we measure lipid levels as a proxy for certain types of heart health (and people are trying to develop new biomarkers for cardiac health). For most diseases, we do not have any form of biomarker from blood, saliva, or anywhere else that would be easy to track and use to expedite drug discovery. Given all the data we could theoretically generate per person, and all the ML/AI algorithms and approaches, it is a bit shocking that biomarkers are in such a primitive state.

Novel biomarkers. Develop novel biomarkers for disease states as a way to expedite drug development and study disease course. This may not be great as a stand alone company, unless some of the biomarkers are good replacements for less comfortable procedures. For example, Exact Biosciences is a $10 billion market cap company, as its product lets you do a biomarker test of your poop for colon cancer risk versus the less comfortable colonoscopy. One could imagine many more of these sorts of tests.

Smell / volatile molecules. Dogs can smell if people have certain types of cancer. This means some biomarker(s) of cancer is emitted into the air and detectable by a dog nose. Why don’t we have similar sensors built? One example in this direction to screen lung cancer.

Other health areas.

There are a wide variety of other things that may impact human health that go ignored - for example air pollution levels and cognitive function - that may have broader societal value. Those may be subject of a future post.

NOTES

[1] There are one off companies in each area. In some cases good ones, in other cases the translational biotech market seems to not have any legitimate players.

[2] References included are not the canonical or key papers in many cases - rather just wanted to show evidence these things are possible. Full scientific citing would take a lot of time and I am a bit overloaded.

[3] I am generally not investing in much biotech for the last few years. However, if there was an incredibly compelling team taking a smart approach in this area I would be interested.

Market Ending Moves

Elad Gil — Tue, 08 Apr 2025 18:26:50 GMT

Every once and a while, there is a move that end questions and competition for startups in a market. A single move can effectively win the market, or “end the market”.

Market ending moves may include:

Merging with your main competitor to remove pricing pressure and options from buyers. The merger between X.com and Paypal in the 1990s consolidated the main payment providers on the internet at the time. Uber and Lyft are rumored to have almost merged at the height of their competition in the 2010s. The rumor is Uber walked from this deal, but then went on to merge its subsidiaries with the major player in multiple international markets to increase leverage of the combined player in the market (China, Russia, etc). Private to private mergers, when companies are still young, are much easier to get past regulators than public/private buys. There are other dynamics to contend with however around ownership, leadership, and ego[1]
Buying a key supplier (of unique data, a bespoke sensor or component, etc) to starve others of a key input to products, prioritize yourself for volume, or get a cost advantage
Key distribution deal. IBM distributing Microsoft’s early O/S, or Yahoo! distributing Google were two “king making” moves in an industry.
Destroy a competitors cash cow. Sometimes your incumbent competitor will have a legacy cash cow business the funds everything else. Offering a free or cheap version of this, or changing business models to destroy their cash flow can be quite effective. This is part of the concern for Google in search/ads relative to some genAI products[2].
Capital. Sometimes you can raise an enormous sum to buy distribution or saturate a network effect market. Tiktok notoriously bought traffic early on at large scale, and Google invested billions in early search distribution. One could argue the current SOTA LLM models are reasonably locked into an oligopoly market due to ability to raise billions or tends of billions of dollars for the next giant model.
Other moves. Lots of other things can be done as well.

When you are thinking about market ending moves, be creative! You can brainstorm literally any scenario. Can you convince a large public company to spin out a key subsidiary to merge with you? Can you put aside ego with your main competitor to combine forces and stop competing for everything? Think broadly. Even if doesn’t happen, this often sparks key thinking on M&A, partnerships, and key hires that you may not have considered otherwise.

Notes

[1] Private to private mergers are notoriously hard between fierce competitors. Usually three topics come up - (a) relative ownership post merger, and (b) who is in charge (c) hard feelings between founders if one felt the other copied them, trash talked them, or the like.

[2] I think Google is most likely to transition just fine (ie survivable vs existential), but many worry this will be an issue

Leaving Delaware

Elad Gil — Wed, 05 Feb 2025 18:22:06 GMT

Delaware, with a population of less than 1 million people, is home to over 1.8 million business entities, including more than 60% of Fortune 500 companies. The state was traditionally viewed as having a business-friendly legal structure and specialized business court system (the Court of Chancery) which made it a preferred incorporation destination.

Businesses seek environments with clear legal frameworks and precedent, low judicial activism, and a business friendly environment. If the environment feels more arbitrary, litigious, or politicized in nature, it can harm the ability for a business to serve its customers and to function. If given a choice, business may avoid or leave these jurisdictions.

Recently, Delaware has had a series of lawsuits, “creative writing” in case law and politicized judges that has been changing the perception of the state as a stable place for companies to legally base their businesses. Companies like SpaceX, Tesla, Dropbox, and others have either left, or disclosed plans to move their incorporation from the state of DE. The primary states companies are considering for the move include Montana, Wyoming, Nevada, and Texas, with most eventually moving to either Nevada or Texas.

This post breaks down some of the pros and cons of Nevada, Texas, and Delaware as well as discusses high level process to leave the state. TradeDesk published a good primer as well.

While many companies have physically moved their headquarters from California to other states (Palantir, Oracle, SpaceX, to name a few) most CA-based companies have left their incorporation in Delaware. This is starting to shift as more companies consider moving Incorporation to Nevada or Texas.

Examples of problems in Delaware causing companies to leave

Part of the concern in Delaware is, as a WSJ opinion piece recently put it, “the likelihood of expensive, meritless or value-destroying litigation leads public companies in Delaware to avoid deals they would otherwise make.” The other part is a few more extreme Delaware activist judges who apply “creative writing” to interpreting the law. The most famous example of this is a court invalidating Elon Musk's board and then subsequently a shareholder-approved compensation package. This ruling was viewed as political activism by many versus following any form of legal framework. Tesla subsequently reincorporated to Texas with 84% non-controller shareholder support.

However, there are other recent examples beyond Tesla. These include:

Trade Desk - Delaware court required special committee approval for controller transactions that were fully disclosed at IPO and had operated without challenge for years
Activision - Delaware court invalidated merger approval process on technical grounds despite 98% shareholder approval and no evidence of harm
Moelis & Company - Delaware court invalidated long-standing stockholder agreements between the company and its founder that were disclosed in IPO; required legislative fix
TripAdvisor - Delaware court ruled that merely reincorporating to Nevada required special committee approval and minority shareholder vote because reducing litigation exposure was deemed a "non-ratable benefit" to the controller

Three articles on why companies are leaving Delaware.

State by State Comparisons from Ben Potter & Team at Latham, Watkins LLC.

Most analysis of which state to incorporate in to collapses down to a choice between Delaware, Nevada and Texas. You can find a full analysis from Ben Potter & Team at Latham here (they have kindly open sourced a quick analysis they did for me and my team):

Latham State Of Incorporation Comparison

1.26MB ∙ PDF file

Download

In general, Delaware has been historically seen to be stronger on:

Historical development & predictability of corporate law (although judicial activism is impacting this)
Experience of judiciary (specialized corporate law judges)
Office of the Secretary of State (ease of set up and doing business)

While Nevada and Texas have been better on:

Litigation risk
Business judgment rules
Takeover defenses
Director and Officer indemnification & exculpation
Expenses of doing business

Between NV and TX, there seem to be slight tradeoffs between each state enumerated below. One plus for NV is that fiduciary duties are established in statute rather than case law (Delaware and Texas being in the case law camp), so theoretically Texas could be subject to the same risk of politicization / creative writing over time that is leading companies to exit Delaware. The hope is that Nevada’s statute-based approach will offer more predictability, which was historically the draw for Delaware. However there are obvious benefits to Texas as well, leading companies like Tesla and SpaceX to decide to move there.

Who is moving where?

Companies that have left, or announced, plans to leave Delaware for:

Nevada: Dropbox, TripAdvisor, Neuralink, Pershing Square, and others
Texas: SpaceX, Tesla, and others

Tradeoffs DE, NV, TX

Tradeoffs to consider in terms of where to incorporate include case law, judges, fiduciary duties and other interests, as well as expense and cost. See also a write up from TradeDesk.

Full analysis from Ben Potter & Team at Latham here:

Latham State Of Incorporation Comparison

1.26MB ∙ PDF file

Download

On the margin it seems a few more companies are choosing Nevada over Texas, although obviously SpaceX and Tesla are giant companies who decided to move to Texas. Anecdotally it seems like more corporate entities have moved to Nevada, while more market cap has moved to Texas.

Legal, governmental, and statutory considerations

Expenses

The other tradeoff faced by companies is the cost of doing business in each state. Expenses seem possible lower in Nevada or Texas versus Delaware when comparing the following:

How to leave Delaware

If you are just incorporating a company now, it is worth considering Nevada or Texas up front versus dealing with it later.

If you are a private company, moving to Nevada or Texas is straightforward. It will probably take a few months start to finish, but with most of the lift done by lawyers and accountants working in the background and not require much of your time as CEO. This process is known as domestication or conversion, depending on the state.

You will have to do a shareholder vote and/or board approval, initiate a conversion/domestication filing in both DE and NV/TX, obtain formation/articles of incorporation and pay a filing fee. You lawyers will then draft new bylaws compliant with new target state laws, revise or update any stock certificates, etc. Finally you may get a new EIN, file final returns in DE and initiate in the new state, and update contracts, employee benefits etc. There may be some industries with special licensing, permitting, or professional accreditation needs, but these tend to be manageable. It is basically a bunch of paperwork, but not too onerous.

If you are a public company seeking to leave Delaware, the most important thing to focus on is proper process. Process is impossible to fix after the fact, and Delaware prefers companies not leave (it is a major source of state revenue). Adhering to good process up front allows you to avoid Delaware court action or unnecessary litigation. This was an issue for TripAdvisor as it tried to leave the state for Nevada.

For a public company, the first step is to create a special committee of independent directors to start and hold the conversation on where to domicile the company. The committee needs to take a broad view of the potential move and re-incorportation and whether it benefits all shareholders (and not just the controller and/or directors of the company - which is what happened in the TripAdvisor case). Things that benefit all shareholders may include lowering costs of doing business, decreased litigation risk, or related. Everything needs to be cleanly documented and kept objective.

If you have 50% of more voting control of your company, you can initiate the move via board approval & written consent. If not, you will likely need to go down the full shareholder vote route. This opens up the potential for more complexity not only in terms of e.g. proxy filings, but also increased potential rejection by minority shareholders. This will not block the move entirely but will increase litigation risk quite a bit.

TradeDesk has a good write up on some aspects of this.

Politically exposed and founder-controlled companies may be the first to move

Given recent activism in Delaware courts, there are two types of companies most likely to leave the state and convert to Nevada or Texas entities. This includes founder-controlled companies that want more leeway and protection in how they operate, and politically exposed companies. Political exposure may include who the owner is (Elon Musk for example), what is the nature of the business, and its perception as politically relevant to be made an example out of. Undoubtedly this will be a minority of companies (right now >60% of the Fortune 500 are in Delaware). Just as a subset of companies has been trickling their headquarters out of California due to its high tax rates, extra regulatory burdens, and poor governance (see Palantir, Oracle, HP, Tesla, and others), one could expect an increasing flow of companies moving their incorporation out of Delaware over time in parallel.

The hope, of course, is that Delaware is able to correct some of the issues it has had in the recent past in order to maintain its position as the default for company incorporation. In the meantime, more companies are actively considering a move.

Discussion w Arthur Mensch, CEO of Mistral AI

Elad Gil — Fri, 22 Mar 2024 12:19:30 GMT

Topics covered:

Mistral
Open and closed source AI
Future tech (small models, context windows, etc)
EU AI & startup scene
Enterprise AI needs
Building fast moving teams

Video link:

Transcript:

DYLAN FIELD

Hi everybody. Welcome. Thank you so much for being here. I am so glad that we're able to host this at Figma. My name is Dylan Field. I'm the CEO and co founder of Figma. And a big welcome to everybody here, but also everyone who's joining us via live stream as well. And I'm really excited for tonight. I think this is going to be a pretty credible conversation and I'm proud to be able to introduce the two folks who'll be having it.

So first, Elad Gil. Elad is not only a dear friend and mentor of mine, but also to many in Silicon Valley and the startup community globally, and also Arthur Mench. Arthur is a former academic who has turned CEO and co founder of Mistral. And mistral for the one or two people in the room that do not know, is breaking incredible ground in open source models and I would dare say changing quite a lot about the future of AI.

And with that, I'll pass it off for their fireside. Welcome.

ELAD GIL

Oh, thanks. Thanks so much to Figma for hosting us, and thanks everybody for making it today. And of course to Arthur. Arthur made a heroic effort to join us where he literally had to jump out into traffic, grab a bike and bike over here. So thank you so much for coming.

ARTHUR MENSCH

Discovering the US, I guess.

ELAD GIL

So from a background perspective, you got your PhD in machine learning, you were a staff research scientist at DeepMind, and then you started Mistrall, and you started it, I believe, with both some folks from Google, such as yourself, and then some folks from meta and the llama project there. You folks have taken an open core approach, which I think is super interesting and we can talk about in a little bit. But I was just curious to start off, what was the impetus for starting? Mistral All, how did you decide to do it? What were the motivations and the initial formation of the company?

ARTHUR MENSCH

Yeah, so I think this has always been on the mind of me, Guillome and Timothe. So I was at DeepMind, they were at meta, and I guess we were waiting for the hour, and the hour came with GPT to some extent, so that we realized we had an opportunity to create a company pretty quickly with a good team that we could hire from day one and go and try to do speedrun a bit because we weren't starting the first. So that's how we got started.

ELAD GIL

I guess the people who are probably watching the live stream, I think the people in the audience are probably well versed with what Mistral does. Can you explain a little bit about the set of products you have the platform, all the various components now.

ARTHUR MENSCH

Yeah, for sure. So Mistral is actually a company building foundational models. We are the leading in open source models. So we have started the company by creating text to text generation models, which are really the foundational block for creating today's generative VA applications. I know we're at Figma, so we're not focusing on images yet, but this is obviously coming at some point. And so, yeah, the differentiation we have is that we took this open core approach to release Mistral 7B mixed hole 87 B in December and build a platform on top of these open source models with the addition of commercial models that we introduced in December and then in February. So we're building an open source models and we're building a portable platform for enterprises with focusing on the developers and building tools for developers.

ELAD GIL

How long did it take from when you founded the company to when you.

ARTHUR MENSCH

Launched 7B took four months, approximately.

ELAD GIL

Yeah. That's amazing. So I think one of the things that's really noticeable is the immense speed in terms of how rapidly Mistral actually launched its very first product and then the rapid adoption of that right as 7B came out suddenly, I think people realized that you could have these small performant models that were very fast. Inference time and time to first token were very cheap, which made a big difference if you were doing things at high throughput. How did you build something so rapidly? Or how did you focus a team on such a singular goal so quickly?

ARTHUR MENSCH

Well, I guess we thought about what was missing in the field and we realized that small models were actually quite compelling for people. We saw a community building on top of llama at the time, on top of llama 7B. But llama 7B wasn't good enough. And so we realized that we could make it much better. We could make a 7B model much better. And so that's the sweet spot we targeted for our introduction to the world. And basically we had to build the entire stack from scratch. So getting the data, building the training code, getting the compute, which was a bit of a challenge because in these four months we were ramping up. So we started at zero GPUs and we actually trained on like 500 GPUs, 7B, I guess we went fast because the team was very motivated and so not a lot of holidays during these four months. And generally speaking, AI teams that succeed and go are typically like four to five people. And AI teams that invent things have always been this size. So we are trying to have an organization where we have squads of five people working on data, working on pretraining, and so far this has worked out quite well.

ELAD GIL

Is there anything you can share in terms of what's coming next in your roadmap?

ARTHUR MENSCH

Yeah, so we have new open source models, both generalist and focused on specific verticals. So this is coming soon. We are introducing some new fine tuning features to the platform and we have introduced a chat based assistant called the Shah that is currently just using the model. So it's pretty raw. It's a bit like chat GBT V zero, and we're actively building on building data connectors and ways to enrich it to make it a compelling solution for enterprises.

ELAD GIL

What kind of verticals do you plan to focus on, or can you share that yet?

ARTHUR MENSCH

Well, I guess we started with financial services because that's where most of the maturity was. Basically we have two go to markets. So enterprises starting with financial services because they are mature enough, and the digital native. So talking to developers like building AI companies or introducing AI to formerly non AI companies, and so that's the two, I guess, go to market pools that we're talking to. The first one through some partnerships with cloud because as it turns out, they're a bit controlling the market in that respect. And then through our platform, we're talking directly to developers.

ELAD GIL

I guess on the cloud side, one of the relationships you recently announced was with Microsoft and Azure. Is there anything you can say there about that relationship or that access that it's providing you to the enterprise?

ARTHUR MENSCH

Yes, this opened up new customers. A lot of enterprises can't really use third party SaaS providers easily because you need to go through procurement, risk assessment, et cetera. But if you go as a third party provider through the cloud, you actually get an accelerator. And so when we believed Mistral Large on Azure, we got like 1000 customers pretty right away. The truth is you need to adapt to the fact that enterprises are using cloud and they don't want to introduce new platforms easily. And so you need to go through that, at least at the beginning.

ELAD GIL

And then one of the things that a lot of the industries focus on right now is scaling up models and ever larger, ever more performant versions. How do you think about the scale that you all are shooting for in the next six months or year? Or is the plan to have very large models over time? Or how do you think about the mix of things that you want to offer?

ARTHUR MENSCH

Yeah, so we first focused on efficiency to be able to train models more efficiently than what was currently done. And then once we had achieved this efficiency, we started to scale so that's why we did another fundraising, and that's why we started to increase the amount of compute we had. And so we can expect new models that will be more powerful because we are pouring more computes in it, models that might be a bit larger, because when you grow the compute, you need to increase the capacity of models. But something that remains very important for us is to be super efficient at inference and to have models that are very compressed. And so that's the kind of model that will continue shipping, especially to the open source world.

ELAD GIL

One of the things that was pointed out to me that I'd love to get your views on is that as you reach certain capabilities within a model, you can start to accelerate the pace at which you can build the next model, because you can use, say, a GPT four level model to do rlaif, or to generate synthetic data, or to do other things that really accelerate what you're doing. Data labeling, all sorts of things, in some cases superhuman performance. How do you think about using models to bootstrap each other up? And does that actually accelerate the timeline for each subsequent release?

ARTHUR MENSCH

Yeah, I guess. Generally speaking, two years ago, RLHF was very important. Today it's actually less important because the models have become better, and they're actually sometimes good enough to self supervise themselves. And what we have noticed is that we scale as we scale. This is definitely improving. So that means that the costly part of going through human annotations is actually reducing. And so this is also lowering the barrier of entrance.

ELAD GIL

I guess another sort of adjacent area is reasoning. And a lot of people feel that as you scale up models, they'll naturally acquire reasoning. And then there's other approaches and entire companies that have recently been founded around just focusing on the reasoning aspect of some of these models. How do you think about that? Are you going to be training sub models for reasoning, or do you think it's just going to come out of scaling the existing models? Is it a mix of the two?

ARTHUR MENSCH

Well, I guess reasoning comes from. At this point, the only validated way of improving reasoning is to train models on larger data and make them bigger. There's obviously some possibilities that you have by building an auto loop, adding new function, calling, adding data as well for the model to reason about grounded aspects instead of trying to imagine stuff. So I guess we don't pretend to have like a secret recipe for reasoning, but we've made models that are pretty good at reasoning by focusing on the data. We're pretty good at using mathematics in our data. And so that's a good way of improving reasoning. There's many ways in which to improve it. Code has helped as well, and so there's no magic recipe, but just focusing on the little things makes it work.

ELAD GIL

Yeah, I guess one of the reasons I ask is, I feel like if you look at sort of the world of AI, there's a few different approaches that have been done in the past. One is sort of the transformer based models and scaling them. The other is a little bit more in the lines of like, Alphago and poker and some of the gaming related approaches where you're doing self play as a way to bootstrap new strategies or new capabilities. And those are in some sense forms of reasoning. And I know that there are certain areas where that may be very natural to do in the context of model training. Code would be an example. There's a few others where you can sort of test things against real rubric. And so I don't know if you folks are considering things like that, or if that's important or not in your mind.

ARTHUR MENSCH

So Guillaume and Timote were doing theorem proving with LLMs back in the day at meta. So that's very linked to, well, using LLM as the reasoning brick and then building an auto loop that involves sampling, that involve Monte Carlo research, all these kind of things. I think the one thing that was standing in the way for this is the fact that models have very high latency, and if you want to sample heavily, you need to make them smaller. And so it's very much tied to efficiency. So as we grow efficiency, as hardware increases in capacity as well, you become more able to explore more and to sample more. And that's a good way effectively to increase reasoning through the autoloup development.

ELAD GIL

And then I guess the other thing a lot more people are talking about or thinking about is memory and some ability to maintain a longer view of state in different ways across actions or chaining things for agents. Do you expect to go down any sort of agentic routes anytime soon? Or is the focus much more on sort of core APIs that are enabling in all sorts of ways?

ARTHUR MENSCH

So that's what we started to enable with function calling, which is a good way to handle, to start creating agent that store states. So memory, when we talk about memory, like of conversation, the way you make it happen is that you basically introduce some crude functions on your middleware part that you give to the model, and so it can actually use that to update its memory and its representation. And so function calling is the one multipurpose tool that you can use to create complex agent. It's hard to make it work, it's hard to evaluate as well. So I think this is going to be one of the biggest challenge. How do you make agent that work, evaluate them and make them work better for feedback? And this is one of the challenge that we'd like to tackle on the product side.

ELAD GIL

And then I guess the one other thing that a lot of people have been talking about recently is just context window. And for example, I know that there's some recent results around, for example, biology models, where if you increase the context window, you can end up with better protein folding and things like that. So the context and the context really matters. I think Gemini launched a million, up to a few million sort of context window, and then magic, I think, has had 5 million for a while. How important do you think that is? Do you think that displaces other things like rag or fine tuning? Are all these things going to work coincident with each other?

ARTHUR MENSCH

So it doesn't displace fine tuning because fine tuning has a very different purpose of pouring your preferences and basically demonstrating the task. On the other hand, it simplifies rag approaches because you can pour more knowledge into the context. And so what we hear from users is that it's like a drug. So once you start to use models with a large context, you don't want to go back. And so that's effectively something we want to try to improve and extend. There's a few techniques for making it happen. On the infrastructure side, it's actually quite a challenge because you need to handle very large attention matrices, but there are ways around it.

ELAD GIL

I see what you're saying. So basically, like on the RAM, on the GPU, you basically ran out of space for something as you're building a bigger and bigger context window. Or is it something else?

ARTHUR MENSCH

Yeah, there's a variety of techniques you need to rethink for sharding and communication to handle the big matrices. And then you do pay a cost because it basically becomes slower because of the quality cost.

ELAD GIL

When do you think we hit a moment where these models are better than humans at most white collar tasks? Do you think that's two years away, five years away, ten years away?

ARTHUR MENSCH

I guess it depends on the task. There's already a few tasks on which the model are actually better. And so I expect this to unfold pretty quickly, actually. So hard to say a date, but I would say in three years this is going to look very different, especially if we find a way to deploy agent and to evaluate them and make them robust and reliable.

ELAD GIL

What about displacing the CEO of Figma? No, I'm just kidding. Just kidding. Dylan, please keep us. So I guess there's a lot of different foundation models that people are starting to work on, right? There's obviously a lot of attention on the LLMs, and there have been diffusion models for image gen, although it seems like people are moving more and more towards image or transformer based approaches for image and video and other things. Are there big holes in terms of where you think there are gaps where people aren't building foundation models, but they should be?

ARTHUR MENSCH

I would say we've seen some things happening on the robotic side, but I think it's still at the very early stage on the audio. This is covered on video. This is starting to be covered, essentially, like models that can take actions and become very good at taking actions. I don't think this is very well covered. There's some progress to be made there, but, yeah, overall, I expect all of this to converge towards similar architectures and at the end of the day, like a joint training as we move forward in time.

ELAD GIL

So do you think eventually everything is a transformer based model?

ARTHUR MENSCH

Well, transformer are a very good way of representing associations between tokens or between information, so it really does not really matter, but it seems to be enough, it seems to be a sufficient representation to capture most of the thing we want to capture, and we know how to train them well so we can transfer information between what we learn from text on images, et cetera. And so that's why I think this is going to be quite hard to displace.

ELAD GIL

Do you think that'll also apply to the hard sciences? If you're trying to do, like, physics simulation, material sciences, pure math.

ARTHUR MENSCH

I don't expect just next token prediction to solve that. And so you do need to move to the outer loop, and you need to figure out also a way to make models interact with simulators, potentially, because at some point, you need the model to learn the physics, and so you need to bootstrap that with the simulator. But I'm not an expert, to be honest.

ELAD GIL

And then all these models, of course, need a lot of GPU, and people have very publicly talked about how there's a GPU crunch right now, and there's shortages of different sorts. When do you think that goes away, or do you think that goes away?

ARTHUR MENSCH

So I think that probably eases as the H and rit comes, and we'll start to see some competition on the hardware space, which is going to improve cost, I think. I expect that also as we move to foundational models that are multimodal, et cetera, we can actually train on more flops. And so I don't think we haven't hit the wall there in scaling. And so I expect this is probably going to continue on the training part and on the inference part as we move on to production and we have models running agent on the background. So really removing this bottleneck that we had at the beginning, which was the speed at which we could read information, I expect that inference capacity will spread pretty significantly.

ELAD GIL

Do you think that will be done through traditional GPU based approaches, or do you think we'll start having more and more custom asics, either for specific transformer models where you burn the weights on the silicon, or more generally for transformers in general, where you can just load a set of weights or something.

ARTHUR MENSCH

So the good thing about the fact that everybody is using transformer is that you can specialize hardware to this architecture and you can make a lot of gains there. There's a few unfortunate bottleneck on Nvidia chips, for instance, the memory bandwidth is a problem. And so by moving on to more custom chips, you can reduce significantly the cost of inference. It's not really ready yet, so we're not betting on it right now, but I really expect that this is going to improve cost pretty significantly.

ELAD GIL

So mistral really started off as a developer centric product, right? You launched to something that was very open source. Now you're starting to serve a variety of enterprises. Is there any commonality in terms of the types of use cases that people are coming with or the areas that enterprises are most quickly adopting these sorts of technologies or approaches?

ARTHUR MENSCH

Yeah. So enterprises adopts the technology for mostly three use cases. So the first one is developer productivity. And usually they kind of struggle with the off the shelf approach because it's not feed to their way of making, of developing. They also use knowledge management tools, and usually they've built their own assistant connected to their database. And the last thing is customer service. So the most mature company have made a large progress toward reducing their human engagement with customers and just making it much more efficient. And so these are really the free use cases we see with enterprises. And with AI companies, it's much more diverse because they are a bit more creative. But yeah, overall, enterprises have these free use cases. It's also the reason why we are starting to think of moving a bit on the value chain and offer things that are a bit more turnkey, because sometimes they need a little bit of help.

ELAD GIL

Yeah, that makes sense. I'm guessing many people here saw the tweet from the CEO of Klarna where he's talking about customer success and how they added a series of tools based on top of OpenAI that basically reduced the number of people they needed by 700. In terms of customer support. They launched it in a month and they had 2.3 million responses in that single month. So it seems like there's this really big wave coming that I think is almost under discussed in terms of impact of productivity, impact of jobs and things like that.

ARTHUR MENSCH

Yeah, so we saw even more diverse use cases. One of them was having a platform that engaged with temporary workers to try and find a job for them. So through texting, and so the customer in question went from 150 people activating this well, engaging directly with customers to seven, and they were actually able to scale the platform much more and to enable temporary workers to work more easily. And generally speaking, this approach of automating more of the customer service is a way to improve the customer service. And so that's, I think, what is exciting about this technology.

ELAD GIL

What do you think is missing right now, or what is preventing enterprise adoption from accelerating further?

ARTHUR MENSCH

So our bet is that they still struggle a bit to evaluate and to figure out how to verify that the model can actually be put in production. What's missing are a bunch of tools to do continuous integration, also tools to automatically improve whatever use case the LLM is used for. And so I think this is what is missing for developer adoption within enterprises. Now for user adoption within enterprises, I think we're still pretty far away from creating assistant that follows instruction, well that can be customized easily by users. And so yeah, on the user side, I think this is what is missing.

ELAD GIL

One thing that I think you've been very thoughtful about is how to approach AI regulation. And I know that you've been involved with some of the conversations in terms of EU regulation and other regulation of AI. Could you explain your viewpoint in terms of what's important to focus on today versus in the future and how to think about it more generally?

ARTHUR MENSCH

Yeah, so we had to speak up because at the time, in October, there was a big movement against open source AI. And so we had to explain that this was actually the right way to today, make the technology secure and well evaluated. And overall we've been continuously saying that we're merging very different conversations about existential risk, which is ill defined and that has little scientific evidence for this is merged with a discussion about, I guess, national security and AIA and LLMs being used to generate bioweapons. But again, this is something that is lacking evidence. And then there's a bunch of very important problems that we should be focusing on, which is how do you actually deploy models and control what they are saying? How do you handle biases? How do you set the editorial tone of a model in a way that you can evaluate and control? And I think this is the most important part. How do you build safe products that you can control well and that you can evaluate well? And this is the one thing we should be focusing on. That's what we've been saying for a couple of months because we were a bit forced to speak up.

ELAD GIL

Yeah, it seems like one of the areas that people are kind of worried about in the short term on AI is things like deepfakes or people spoofing voices or other things like that, either for financial attacks, for political purposes, et cetera. Do you all have plans to go down the voice and sort of multimodality side?

ARTHUR MENSCH

So generating things that are not text is effectively a bit more of a trap on the safety side, and that we've avoided it effectively. Imitating voices and deepfakes are very concerning. And this is not something that we pretend to be able to sort text. It's much easier because there's never this kind of problem. So you can generate text is generating text is never an enabler of very harmful behavior. Misinformation has been mentioned, but usually misinformation is bottlenecked by diffusion and not by creation. So by focusing on text, we kind of circumvent these issues, which are very real.

ELAD GIL

I think one of the things that's very striking about Mistral is, and I should say in Europe in general right now, is there's a very robust startup scene. And if I look at the two biggest pockets of AI right now in terms of startup formation, it's basically here in Silicon Valley, and then it's like the Paris London corridor, and you have eleven labs and you have Mastrall and you have all these great companies forming. What do you think is driving that?

ARTHUR MENSCH

I think there's a couple of historical reasons. In London there was, and there still is DeepMind, which was a very strong attractor of talents across the world. And in Paris in 2018, both DeepMind and Google opened offices, research offices, and it went and augmented the existing research scene that was already pretty strong, because as it turns out, France and also a couple of other countries in the European Union have very good education pipeline. And so junior machine learning engineers and junior machine learning scientists are quite good. And so that's one of the reason why today we have a pretty strong ecosystem of companies on both the foundational layer, but also on the application layer.

ELAD GIL

Yeah, the French seem a lot smarter than the British. So. No, I'm just kidding.

ARTHUR MENSCH

I'm not the one saying that.

ELAD GIL

The other thing that I think is kind of striking is you start to see a lot of different AI based companies focused on regional differences. So, for example, when you launched, you included a variety of different european languages. I know there's models being built right now for Japan, for India, for a variety of different geos. And one could argue that either you have large global platform companies that serve everywhere, except for maybe China, because China is likely to be firewalled in some ways, just like it has been for the Internet more generally. Or you could imagine a world where you have regional champions emerge. And in particular, you could almost view it like Boeing versus Airbus, where the governments of specific regions decide that they really want to fund or become customers to local players. What do you view as sort of the future world, and how does that evolve in terms of global versus regional platforms?

ARTHUR MENSCH

So we've taken a global approach to distribution. I guess there was another path that we could have taken, which was to focus on just the european market, pretending that there was any form of defensibility there. We don't think this is the case. Technology remains very fluid and so can circulate across countries. On the other hand, the technology we're building is effectively very linked to language, and language is, well, English is only one language across many. And as it turns out, LLMs are much better at English than other languages. So by also focusing more on different languages, we managed to make models that are very good at european languages in particular, versus the american models. And so there's a big market for that. And similarly, there's a big market in Asia for models that can speak asian languages. And there's a variety of scientific problems to be sorted and solved to address these markets, but those are huge, and those haven't been the focus of us companies. So it's effectively an opportunity for us as a european company to focus a bit more on the world globally.

ELAD GIL

Okay, great. I think we can open up to a few questions from the audience, and if people want to just ask, I can always just repeat it in the back there, please. Yeah, right there. If you want to speak loudly, I can repeat what you say. The question is, do you plan to release closed source versions of your model or always be open source?

ARTHUR MENSCH

So we have commercial models already. So to an extent we haven't been open sourcing everything. We are a very young company, but our purpose is to release the best open source models. And then we are basically coming up with an enterprise surrounding and some premium features that we can sell to sustain the business. And so our strategy today, and that might evolve with time, is to have both very strong open source models, but also models that are much stronger actually at that point in time as closed source APIs. The one thing that we focus on also for our commercial models is to make deployment of these models very portable and very flexible. So we have customers to whom we ship the weights and allow them to modify the model, do client side fine tuning the same way they would do it with open source models. And so in that sense, we have some coherence across the commercial family and the open source family.

[AUDIENCE QUESTION ON MAIN USES CASES]

ARTHUR MENSCH

Knowledge management, developer productivity. So coding basically.

[AUDIENCE QUESTION – PLANS TO DO CODING SPECIFIC MODELS?]

ARTHUR MENSCH

Yeah, we have plans. Not doing any announcement today, but we do have plans.

[AUDIENCE QUESTION – NEW ARCHITECTURES AND RESEARCH]

ARTHUR MENSCH

We've been mostly into production at that point because the team was pretty lean, but we're not dedicating a couple of full time employees like finding new architectures, finding well, making research. And I think this is super important to remain relevant. So as we scale, we will be able to afford more exploration. It's also very linked to the compute capacity you have. So if you want to make some discoveries and make some progress, you need to have enough compute. And we're a bit compute bound because of the shortage on H 100, but this is going to improve favorably. So we expect to be doing more research and more exploratory research, I guess because we've been doing research from the.

ELAD GIL

Starter, I guess related to that, it seems like in general, your team has a very strong bias for action, and you move very quickly. How do you select for that in people that you hire? Are there specific things you look for, interview questions you ask?

ARTHUR MENSCH

So we look for AI scientists that can do everything from going down the infrastructure stack to making, extract, transform and load pipelines to thinking about mathematics. So we've been trying to find full stack AI engineers, and they tend to have a strong bias for action. Really, the focus we had is to find low ego people willing to get their hands dirty with jobs that are considered boring by some AI scientists because it's a bit boring. But this has been actually quite productive. And because we focused on the right things and the back. I guess the team is now quite big, so there's a bunch of challenges associated to that. I was surprised by the amount of inbound that we had and the amount of representation that I had to do, especially as we got drawn into political stuff, which we would rather have avoided, but we kind of didn't have a choice. So this was definitely a surprise for me, generally speaking. I was also surprised by the speed we managed to have, because it actually exceeded our expectations. But, yeah, I had pretty little idea of what the job of a funder would be when we started. It's quite fun, but it's effectively surprising. I was imagining myself as still coding after a year, and it's actually no longer the case, unfortunately. But, yeah, that's the price of trying to scale up pretty quickly.

ELAD GIL

You get to do HR coding now, which is even better.

ARTHUR MENSCH

Yeah.

ARTHUR MENSCH

So the reason why we started the company is to have a production arm that creates fission value, to have a research arm. And to be honest, there isn't much demonstration of existence of such organs, because you do have a few research labs that are tied to cloud companies that have a very big top line and using it to sustain research. We think that with AI and with the value that the technology brings, there is a way for doing it. But I guess this still remains to be shown. And that's the experiment we are making with mistral.

ELAD GIL

Probably. One last question. I know Arthur has a hard stop, maybe way in the back there.

[AUDIENCE QUESTION – HOW MUCH PERFORMANCE CAN A SMALL MODEL REALLY HAVE]

ARTHUR MENSCH

Yes, I think you can squeeze it to that point. The question is whether you can have a 7B model that beats Mistral Large. This starts to be a bit tricky, but there might be way. I also expect the hardware to improve, like the local hardware to improve. And so that will also give a little more space and a little more memory. And yeah, I see more potential there, because effectively you're a bit constrained by scaling loads. That tells you that at some point you do saturate the capacity of models of a certain size.

ELAD GIL

What is the main constraint? Or what do you think is the thing that it asymptotes against for scaling loads?

ARTHUR MENSCH

You can make 7B models very strong if you focus on a specific task. But if you want to pour all of the knowledge of the world onto 7GB, well, it's actually quite ambitious. So one thing is, for instance, multilingual models at this size are not a great idea. So you do need to focus on a specific part of the human knowledge. You want to compress I guess one.

ELAD GIL

Last question for me and then we can wrap up is a friend of mine pointed this out to me, which basically, if you think about what you do when you're training a model is you spin up a giant data center or supercomputer and then you run it for n weeks or months or however long you decide to train for, and then the output is a file.

ARTHUR MENSCH

You're basically zipping the world knowledge. It's not much more than that, actually.

ELAD GIL

Yeah. How do you think about either forms of continuous training or retraining over time or sort of longer training runs that get tacked on? I know some people are basically training longer or longer and then dropping a model and then they keep training and then they drop a model. And so I don't know how you think about where the world heads.

ARTHUR MENSCH

Yeah, this is an efficient way of training, so that's definitely interesting for us.

ELAD GIL

Okay, great. Well, please join me in thanking Arthur.

Other Firesides & Podcasts

My book: High Growth Handbook. Amazon. Online.

Markets:

Startup life

A brief guide to startup pivots

4 Ways Startups Fail

Founder Investors and Scout Programs

Co-Founders

Raising Money

Things I Don't Know About AI

Elad Gil — Wed, 21 Feb 2024 13:37:01 GMT

In most markets, the more time passes the clearer things become. In generative AI (“AI”), it has been the opposite. The more time passes, the less I think I actually understand.

For each level of the AI stack, I have open questions. I list these out below to stimulate dialog and feedback.

LLM Questions

There are in some sense two types of LLMs - frontier models - at the cutting edge of performance (think GPT-4 vs other models until recently), and everything else. In 2021 I wrote that I thought the frontier models market would collapse over time into an oligopoly market due to the scale of capital needed. In parallel, non-frontier models would more commodity / pricing driven and have a stronger opensource presence (note this was pre-Llama and pre-Mistral launches).

Things seem to be evolving towards the above:

Frontier LLMs are likely to be an oligopoly market. Current contenders include closed source models like OpenAI, Google, Anthropic, and perhaps Grok/X.ai, and Llama (Meta) and Mistral on the open source side. This list may of course change in the coming year or two. Frontier models keep getting more and more expensive to train, while commodity models drop in price each year as performance goes up (for example, it is probably ~5X cheaper to train GPT-3.5 equivalent now than 2 years ago)

As model scale has gotten larger, funding increasingly has been primarily coming from the cloud providers / big tech. For example, Microsoft invested $10B+ in OpenAI, while Anthropic raised $7B between Amazon and Google. NVIDIA is also a big investor in foundation model companies of many types. The venture funding for these companies in contrast is a tiny drop in the ocean in comparison. As frontier model training booms in cost, the emerging funders are largely concentrated amongst big tech companies (typically with strong incentives to fund the area for their own revenue - ie cloud providers or NVIDIA), or nation states wanting to back local champions (see eg UAE and Falcon). This is impacting the market and driving selection of potential winners early.

It is important to note that the scale of investments being made by these cloud providers is dwarfed by actual cloud revenue. For example, Azure from Microsoft generates $25B in revenue a quarter. The ~$10B OpenAI investment by Microsoft is roughly 6 weeks of Azure revenue. AI is having a big impact on Azure revenue revently. Indeed Azure grew 6 percentage points in Q2 2024 from AI - which would put it at an annualized increase of $5-6B (or 50% of its investment in OpenAI! Per year!). Obviously revenue is not net income but this is striking nonetheless, and suggests the big clouds have an economic reason to fund more large scale models over time.

In parallel, Meta has done outstanding work with Llama models and recently announced $20B compute budget, in part to fund massive model training. I posited 18 months ago that an open source sponsor for AI models should emerge, but assumed it would be Amazon or NVIDIA with a lower chance of it being Meta. (Zuckerberg & Yann Lecunn have been visionary here).

Questions on LLMs:

Are cloud providers king-making a handful of players at the frontier and locking in the oligopoly market via the sheer scale of compute/capital they provide? When do cloud providers stop funding new LLM foundation companies versus continuing to fund existing? Cloud providers are easily the biggest funders of foundation models, not venture capitalists. Given they are constrained in M&A due to FTC actions, and the revenue that comes from cloud usage, it is rational for them to do so. This may lead / has led to some distortion of market dynamics. How does this impact the long term economics and market structure for LLMs? Does this mean we will see the end of new frontier LLM companies soon due to a lack of enough capital and talent for new entrants? Or do they keep funding large models hoping some will convert on their clouds to revenue?
Does OSS models flip some of the economics in AI from foundation models to clouds? Does Meta continue to fund OS models? If so, does eg Llama-N catch up to the very frontier? A fully open source model performing at the very frontier of AI has the potential to flip a subportion the economic share of AI infra from LLMs towards cloud and inference providers and decreases revenue away from the other LLM foundation model companies. Again, this is likely an oligopoly market with no singular winner (barring AGI), but has implications on how to think about the relative importance of cloud and infrastructure companies in this market (and of course both can be very important!).
- One of the most brilliant things in the Llama2 terms of use is the open commercial use of the license if you have fewer then 700 million users[1]. This obviously prevents some large competitors from using their models. But it also means if you are a big cloud provider you need to pay a license to Meta for Llama, which Microsoft has already done. This creates an interesting long term way for Meta to control (& monetize) Llama despite being open source.
How do we think about speed and price vs performance for models? One could imagine extremely slow incredibly performant models may be quite valuable if compared to normal human speed to do things. The latest largest Gemini models seem to be heading in this direction with large 1 million+ token context windows a la Magic, which announced a 5 million token window in June 2023. Large context windows and depth of understanding can really change how we think about AI uses and engineering. On the other side of the spectrum, Mistral has shown the value of small, fast and cheap to inference performant models. The 2x2 below suggests a potential segmentation of where models will matter most.

How do architectures for foundation models evolve? Do agentic models with different architectures subsume some of the future potential of LLMs? When do other forms of memory and reasoning come into play?

Do governments back (or direct their purchasing to) regional AI champions? Will national governments differentially spend on local models a la Boeing vs Airbus in aerospace? Do governments want to support models that reflect their local values, languages, etc? Besides cloud providers and global big tech (think also e.g. Alibaba, Rakuten etc) the other big sources of potential capital are countries. There are now great model companies in Europe (e.g. Mistral), Japan, India, UAE, China and other countries. If so, there may be a few multi-billion AI foundation model regional companies created just off of government revenue.
What happens in China? One could anticipate Chinese LLMs to be backed by Tencent, Alibaba, Xiaomi, ByteDance and others investing in big ways into local LLMs companies. China’s government has long used regulatory and literal firewalls to prevent competition from non-Chinese companies and to build local, government supported and censored champions. One interesting thing to note is the trend of Chinese OSS models. Qwen from Alibaba for example has moved higher on the broader LMSYS leaderboards.
What happens with X.ai? Seems like a wild card.
How good does Google get? Google has the compute, scale, talent to make amazing things and is organized and moving fast. Google was always the worlds first AI-first company. Seems like a wild card.

Infra companies

There are a few types of infrastructure companies with very different uses. For example, Braintrust provides eval, prompt playgrounds, logging and proxies to help companies move from “vibe based” analysis of AI to data driven. Scale.ai and others play a key role in data labeling, fine tuning, and other areas. A number of these have open but less existential questions (for example how much of RLHF turns into RLAIF).

The biggest uncertainties and questions in AI infra have to do with the AI Cloud Stack and how it evolves. It seems like there are very different needs between startups and enterprises for AI cloud services. For startups, the new cloud providers and tooling (think Anyscale, Baseten, Modal, Replicate, Together, etc) seem to be taking a useful path resulting in fast adoption and revenue growth.

For enterprises, who tend to have specialized needs, there are some open questions. For example:

Does the current AI cloud companies need to build an on-premise/BYOC/VPN version of their offerings for larger enterprises? It seems like enterprises will optimize for (a) using their existing cloud marketplace credits which they already have budget for, to buy services (b) will be hesitant to round trip out from where their webapp / data is hosted (ie AWS, Azure, GCP) due to latency & performance and (c) will care about security, compliance (FedRAMP, HIPAA etc). The short term startup market for AI cloud may differ from long term enterprise needs.
How much of AI cloud adoption is due to constrained GPU / GPU arb? In the absence of GPU on the main cloud providers companies are scrambling to find sufficient GPU for their needs, accelerating adoption of new startups with their own GPU clouds. One potential strategy NVIDIA could be doing is preferentially allocating GPU to these new providers to decrease bargaining power of hyperscalers and to fragment the market, as well as to accelerate the industry via startups. When does the GPU bottleneck end and how does that impact new AI cloud providers? It seems like an end to GPU shortages on the main clouds would be negative for companies whose only business is GPU cloud, while those with more tools and services should have an easier transition if this were to happen.
How do new AI ASICS like Groq impact AI clouds?
What else gets consolidated into AI clouds? Do they cross sell embeddings & RAG? Continuous updates? Fine tuning? Other services? How does that impact data labelers or others with overlapping offerings? What gets consolidated directly into model providers vs via the clouds?
Which companies in the AI cloud will pursue which business model?
- It is important to note there are really 2 market segments in the AI cloud world (a) startups (b) mid-market and enterprise. It seems likely that “GPU only” business model default works with the startup segment(who have fewer cloud needs), but for large enterprises adoption may be more driven by GPU cloud constraints on major platforms. Do companies providing developer tooling, API endpoints, and/or specialized hardware, or other aspects morph into two other analogous models - (a) “Snowflake/Databricks for AI” model or (b) “Cloudflare for AI”? If so, which ones adopt which model?
How big do the new AI clouds become? As large as Heroku, Digital Ocean, Snowflake, or AWS? What is the size of outcome and utilization scale for this class of company?
How does the AI stack evolve with very long context window models? How do we think about the interplay of context window & prompt engineering, fine tuning, RAG, and inference costs?
How does FTC (and other regulator) prevention of M&A impact this market? There are at least a dozen credible companies building AI cloud related products and services - too many for all of them to be stand alone. How does one think about exits under an administration that is aggressively against tech M&A? Should the AI clouds themselves consolidate amongst themselves to consolidate share and services offered?

Apps questions

ChatGPT was the starting gun for many AI founders. Prior to ChatGPT (and right before that Midjourney and Stable Diffusion) most people in tech were not paying close attention to the Transformer/Diffusion model revolution and dislocation we are now experiencing.

This means that people closest to the model and technology - ie AI researchers and infra engineers - were the first people to leave to start new companies based on this technology. The people farther away from the core model world - many product engineers, designers, and PMs, did not become aware of how important AI is until now.

ChatGPT launched ~15 months ago. If it takes 9-12 months to decide to quit your job, a few months to do it, and a few months to brainstorm an initial idea with a cofounder, we should start to see a wave of app builders showing up now / shortly.

B2B apps. What will be the important companies and markets in the emerging wave of B2B apps? Where will incumbents gain value versus startups? I have a long post on this coming shortly.
Consumer. Arguably a number of the earliest AI products are consumer or “prosumer” - ie used in both personal and business use cases. Apps like ChatGPT, Midjourney, Perplexity and Pika are examples of this. That said, why are there so few consumer builders in the AI ecosystem? Is it purely the time delay mentioned above? It seems like the 2007-2012 social product cohort has aged out. New blood is needed to build the next great wave of AI consumer.
Agents. Lots and lots of things can happen with agents. What will be strong focused product areas versus startups looking for a use case?

This is one of the most exciting and fast-changing moments in technology in my lifetime. It will be fun to see what everyone builds. Looking forward to thoughts on the questions above.

Thanks to Amjad Masad and Vipul Prakash for comments on a draft of this post.

NOTES

[1] Yes I occasionally read terms of use for fun.

MY BOOK
You can order the High Growth Handbook here. Or read it online for free.

OTHER POSTS

Firesides & Podcasts

Markets:

Startup life

A brief guide to startup pivots

4 Ways Startups Fail

Founder Investors and Scout Programs

Co-Founders

Raising Money

A return to company core mission & values

Elad Gil — Wed, 31 Jan 2024 10:32:43 GMT

This post is not about whether a company (or other organizations) should, or should not have, viewpoints on every political and current event. That is ultimately up to the organization’s governance.

However, if you run an organization that is being distracted from its core mission and fatigued on politics at work- now is a good window to let employees know your organization is refocusing on its core mission. Broader politics is no longer part of the day to day at the company, while other performance & cultural values have returned back into focus.

You could potentially tie this to a bigger view point on the nature of work at the company or organization, and a re-affirmation of key cultural values. For example, that you are a performance-based culture, focused on your customers as primary stakeholders of your business, etc. Shopify did this extremely well and gracefully back in 2021.

Why now?

There is a grace period right now between the Israel Oct 7 terror attack and the Biden / Trump election. It is possible 2024 will be a bumpy year, so acting now in relative calm might be the easiest time to make a change. In this window of relatively few, new politically charged events it is a good time to make a statement, that your company will no longer be making statements on every news event. :)

What to let people know?

Your organization’s focus is on your core mission or business. It may be useful to enumerate your mission, some of the core cultural values, and a reminder of which customers you serve, etc.
Your organization is not the government or political adjudicator. Your organization will no longer make political statements on broad based societal topics. It may occasionally comment on areas unique to its specific business. For example, it may be important for a fintech company to engage in financial services policy and regulation, or for a healthcare company to work on healthcare policy.
Your organization will not allow for internal political conversation on official channels (e.g. slack) for political discourse. Not everyone wants to participate and many find this distracting. If you want to have political conversations you are free to do so on your own time with your own personal chats, platform etc.

What else?

People tend to self select into organizations that they will thrive in. If you are explicit in terms of company values through the interview process, all hands, website etc you will select out people who view work as a primary political venue.
A small subset of people may resist change of any sort (be it this one, or any other), and when change happens act out or disrupt work. This may lead to the voluntary, or involuntary, departure of a handful of people.
You may get tested in the next political event. Staying firm and unapologetic tends to work well and prevents slippage of focus and mission. You have told people about the change and now you are sticking with it.
You do not necessarily need to potentially tie this to a bigger view point on the nature of work at the company or organization, and a re-affirmation of key cultural values. If you want you can just send a short note about the shift and mention it at the next all hands. However, it is an opportunity to remind your fellow employees of what mission you serve, or that for example, that you are a performance-based culture, focused on your customers as primary stakeholders, etc. Shopify did this extremely well back in 2021.
For some organizations refocusing back to organizational core mission is a smooth and easy transition. For others there are a few bumpy weeks and then things resolve and life moves on.

MY BOOK
You can order the High Growth Handbook here. Or read it online for free.

OTHER POSTS

Firesides & Podcasts

Markets:

Startup life

A brief guide to startup pivots

4 Ways Startups Fail

Founder Investors and Scout Programs

Co-Founders

Raising Money

Altimeter's Brad Gerstner on Macro, Tech and Startups

Elad Gil — Tue, 05 Dec 2023 16:52:07 GMT

Brad Gerstner, Founder of Altimeter, and I discuss:

What can we learn from tech and internet history?
Technology cycles
AI, its impact, and how to invest in it
Macro shifts
Tech startup world

Video

Transcript

Brad Gerstner

Thanks for having me. I started four companies on the back of a napkin, like, day one. And so I get both excited and know the feeling of sitting where you all are sitting right now. Unfortunately, this guy gets to have all the day-one fun these days because Altimeter tends to invest a little bit later. And I'm sure we'll get into that today.

But, my heart really still lies and the heart rate gets going talking about building new things. Because ultimately what we all do only matters if we're making life better for some consumer or we're making life better for some business, right? Like we're pushing humanity forward or we're not. Everything else is kind of bullshit and financial engineering and so being in the room with the people who are actually doing the work and building the things that keep this engine of innovation going, I couldn't wait to come and spend time with you.

Elad Gil

Yeah, thanks so much for coming. Maybe we can start with your background because you've done a variety of things. You were trained, I think, as a lawyer. You've started multiple companies. You were very early on at venture capital firms like General Catalyst, which have now really taken on a major role in the ecosystem. And then, you started Altimeter in, I believe, 2008.

Brad Gerstner

2008, yeah. So the quick and dirty. I went to law school in 1996. I was always passionate about technology, kind of an early adopter. But I see Andreessen on the cover of, I think, Time magazine in 1996 for the Netscape browser. I gathered all my friends in law school around one of the few computers we had that was connected to the Internet. I pulled up the Netscape browser, and I said, this changes everything. They're like, what are you talking about?

But in that moment, I realized I went to law school. I was trained as an analyst, but what I really wanted to do, where my passion was, was to get to this place, Silicon Valley. I grew up in Indiana. I hadn't traveled a lot. I had never been to California. So I bought a ticket and I flew to Silicon Valley. And that really just started me on this journey. I ended up practicing law, doing this stint in politics, and realized I wanted to go back to business school to really pivot into business. 1999, the world's blowing up. Internet 1.0. I feel like I was already late. I must have missed it. There are already big outcomes.

But of course, we were just getting started. I thought I was going to move to Silicon Valley. In fact, I came out here and interviewed with some small, little companies. I really wanted a small company with a ping pong table and getting that experience. So I interviewed with a company called TellMe - you know, which Mike McCue ran, which was kind of voice automation that Microsoft ended up buying, and then a little search company called Google that I thought was already too big with a few hundred employees, but loved it as a user.

And ultimately, my classmate, who became my wife, said, we're staying in Boston. And I knew these two guys who were trying to start a venture capital firm in Boston, David Fialkow and Joel Cutler, and they said, hey, come work with us. We need to put together a launch deal to launch the fund. And so let's incubate a launch deal, and we need to partner with the best firm and the hottest firm in the world to put ourselves on the map. So the hottest firm at the time was SoftBank. Before they weren't. Before they were. Before they weren't. They've had quite the up and down over the years. But SoftBank in 1999 was flying. They were hosting every party out here.

Everybody was starting an Internet company. We had an idea, David and Joel, really, for an online travel business that basically like Shopify for online travel. Go and build the booking engines, because Expedia was already doing air tickets, but they hadn't done hotels or vacation packages or cruises. So let's go build that booking infrastructure and then we'll license it to the brands. And that's what we did. We pulled the deal together, and within a couple of years, we had a business doing over a billion in gross bookings, and over 20 million in EBITDA, and we would go on to sell that business to Barry Diller in 2001.

I started two additional businesses before getting back to the investing world. And so the red thread for me was definitely a passionate analyst trying to understand and make sense out of the world, but really, as a founder, an entrepreneur, a venture capitalist, walked up and down Sand Hill raising money and had those feels of a first-time founder. All of the exhilaration, all of the terrifying feelings of that. I was running a company with 1,200 people when September 11 happened, and I had to gather them in a room and let 700 people go because the world of travel had stopped in a day.

And so unless you've done that and experienced those things, then I think it's very difficult to have the empathy to sit in the chairs that we now occupy and talk to founders who are on those journeys and so many people along the way. But David and Joel were really instrumental in helping me see what was possible.

Elad Gil

Yeah, it's kind of interesting. You mentioned the feeling of showing up too late. I remember Marc Andreessen talking about this, where he said he showed up in the 90s. The world of shrink wrap software was sort of the prior wave before the Internet. And everybody felt like they were showing up too late to Silicon Valley. Was tech still a thing?

And so I feel like it's almost a generational thing where every generation feels like it's showing up too late and then it turns out in hindsight, there's so much to do and so many opportunities.

What made you decide to start Altimeter in 2008? So to your point, the financial crisis has happened.

Brad Gerstner

I had started these three businesses and to be fair, they were all like base hits, doubles. But for a poor kid from Indiana, they felt like smoking home runs to me. Right. Because you just get economic freedom and it changes how you think. In fact, I remember Joel Cutler saying to me one time, it must have been 99 or 2000, I was fretting about something. He was like, we just need to get money in your bank account so you stop worrying about all the little shit. And I said, I'm happy to send you the wiring instructions. Like, we could solve this right now.

But I knew I wanted by the time I'd started that third business, I knew that I really wanted to run my own thing. And I had some observations about venture capital and about public market investing that were different than what was currently being practiced. So in venture capital, I saw a lot of firms that were very generalist firms. They were organized as traditional partnerships. They moved pretty slow. They were not very diverse. A lot of them at that time were in Boston. I consider myself a founder. Wanted to be in Silicon Valley, wanted to be with my people, wearing my T-shirts. I didn't want to wear suits. And I thought there was a huge advantage from also doing public markets.

So in my early first company, I was lucky enough to have two hedge fund investors invest in that company. Paul Reeder, who ran PAR Capital, and Seth Klarman, who ran Baupost. True legends in the hedge fund industry. Nobody talked about it as like, crossover investing, right? This predated Tiger and Coatue that both got started around that time. But if you go all the way back to Warren Buffett, in 1950, he had been investing in private companies and public companies, right? He commingled them. It really was the emergence of LPs who said, no, you're not allowed to do both. We want you in one of these buckets. That caused firms to start specializing and calling themselves kind of one thing or the other.

I had a theory. Technology companies are going to scale faster because of this thing called the Internet. The private markets and private capital will mature and be able to provide the capital they need to scale faster. And that there was a huge informational advantage from being in both. Right. That I could bring the value of that venture network to the public markets and the public markets to the venture investors. Because if you're bold enough in 2008 to think you're a venture brand that can break in against the Sequoias of the world.

You better have a wedge, like what's your wedge? And you better have a point of competitive differentiation. And so our wedge was really that we bring the knowledge, insights, research of the public markets, the relationship of the public markets to these earlier stage investors, and that we can be with them for a longer period of time. Go and do that Series A with Bill Gurley because he's excellent at helping you build that first two or three years, and find product market fit. Then we'll come in, take the baton, help you scale your capital, still with the sensibility of a founder who wears a T-shirt and lives here and lives and breathes it, but also has a foot in Wall Street, knows how to navigate, kind of scaling that capital. So I thought there was an opportunity.

I was lucky enough to be trained in the hedge fund business by Paul Reader and some of these greats so I understood the public market business. 2008, I certainly didn't think the world was going to be as bad as it was. I had some endowments who had promised me some money at the end of 2007. But this is a good lesson, starting any business, right? So 2007, the world's pretty good. Some people say, hey, we'll give you a couple of hundred million dollars. Go start a fund. We like this idea you have. Public and venture. Move to Silicon Valley. I get married at the end of seven. I have my first child in June of eight. And I had made the plan. I'm going to launch this firm.

And of course, the world starts melting down. And I don't know, entrepreneurs, they're pretty damn determined. And I said the horse has left the stable. We're doing this. I'll get whatever money I get. I had mentors who started with just a little. Seth Klarman had started 1981 with less than $20 million. My mentor, Paul Reader, had started with less than 5 million. So I was just like, the key is you're either committed or you're not. Are you starting or you're not?

And so, November 1, 2008, yesterday was our 15th anniversary. The first trade I ever placed right in my own fund, I bought Priceline was the first stock we bought. The world was melting down. It was hard to even get a technology platform at that time to do our business because everybody thought all the investment banks were going broke. And the guy who gave me my shot at opening up worked at UBS, and he said, we'll be your prime broker. And he said, Listen, here's the truth. I don't know if we're going to be in business in six months, but you're a good guy. Let's give it a shot. And if it works, it'll work together. And he sent me a text yesterday, and it worked beyond all of our wildest dreams.

Elad Gil

Yeah, well, congratulations on 15 years. That's pretty amazing.

Brad Gerstner

Thanks, it’s been a run.

Elad Gil

You folks have also done a variety of things. You've done privates like one of the early runs in Snowflake, you help anchor IPOs, you do the public market investing.

How do you view the range of things that you do today and how's that broken out by relative focus?

Brad Gerstner

Firms are organized in all sorts of different ways. You guys know the classic benchmark model. Six partners, they're each full stack. So they go source their own deals, they do their own deals, they don't have analysts. And the portfolio is just the natural byproduct of the work that they do. And each partner tends to do two or three deals a year.

We have a public fund. So just to give you a sense of the path, we started with less than 5 million. In 2021. We peaked at over 21 billion in assets. We sent 6 billion and change of profits back to our investors - more than all the money we had ever raised from all of our funds combined. And today we're over 10 billion, roughly split equally between our public funds and our venture funds.

On the public side, that capital tends to be what they call evergreen. So somebody will give us their money, we'll go invest it in the public market. We're judged at the end of every year. We have a team of five people and their major is public markets and their minor is the venture deals that we do. So everybody on our team has a major and a minor. We all work in an open setting like this. So we're sharing ideas because again, we think that free flow of information is how we can really understand how these venture insights from the relationships that we have inform whether or not we should be buying or selling Nvidia, right? And obviously, you guys get why that works.

On the venture side, we're now investing out of our 6th venture fund, which is about $1.5 billion. We just had our first close on our 7th. That will be a billion dollars. And there we really think of our sweet spot as being the occasional Series A, but really the sweet spot is that Series B and Series C. So those are in the case of Snowflake, that Series B was $170,000,000. The Series C was 500 million. The Series B, it was still pre-revenue. They had been working on the product for three years. They had ten beta customers. So we had a sense, right, because we could talk to the customers, we could use the product ourselves. We understood why there was going to be product market fit and why rearchitecting the database native for the cloud, separating storage and compute could be so big. So I would say about 90% of the profits we've generated as a firm in venture were in companies that had less than $5 million in revenue at the time we invested.

Elad Gil

When you talk about $170,000,000 for Snowflake, you mean market cap right?

Brad Gerstner

Yeah, that was the total enterprise value of Snowflake when we got involved, and we're still one of their largest shareholders today, and it's over $50 billion in enterprise value. And I think there's a company that will most certainly double and triple from here in the fullness of time.

Let's say you compete for the Series B in Snowflake, and let's say you don't win it. If you are a stage-specific firm, then all the work you've done on that goes on the shelf, and you move on to your next deal. But at Altimeter we may lose the series B. We may choose not to do the series B, that we don't think the risk-reward is a good setup in the series B, but we can do the C. If we don't do the C, we can do the D or the E or we've bought in over 100 IPOs.

We're one of the larger buyers in tech IPOs, so we can buy it in the IPO. And in the case of Snowflake, we bought it along every step along the way because the risk-reward kept getting better. Right. Because although the price was going up, the risk was going down. And so we always think about the distribution of those probabilities, the relative risk and reward when we're making the investment.

So if you ask me, like, Brad, where does your heart really lie? It lies in this room. Right. I'm still the guy who started those companies on the back of a napkin. But the truth of the matter is, we don't have the scale in our organization to be the best partner, who's going to sit there and hire your first engineer and spend that two to three years trying to find product market fit. So I partner with world-class guys like this guy and other world-class investors like Benchmark or Mike Speiser at Sutter Hill or Martin Casado and Andreessen and go through the list, and they're doing those A's, and then we come in and we really want to be the best partner possible to hand the baton to. And then when I can't satiate my own appetite for that early-stage stuff, I come in as an angel investor off my personal balance sheet. And I have, I don't know, far too many angel investments, but I can't resist the excitement in partnering with really world-class people at those early stages.

Elad Gil

So a lot has changed since you started Altimeter in terms of the venture markets, public markets, what do you think are the biggest shifts that have happened in terms of the entrepreneurial ecosystem from 2008 until today?

And what do you think will snap back? Because I feel like there are a lot of changes that happened during ZIRP (Zero interest-rate policy) that happened over the last couple of years. Some of those will keep going and some will shift. I'm curious how you think about this sort of macro side of this.

Brad Gerstner

And I might even rewind a little bit further because I think the pattern recognition of 1999 and then the period 2001, it's super important. So 99, everybody said, I can start a company. This is easy, it's get rich quick. Everything was going to the moon. You can take it public, it has no real revenues, no real profits. That was all make-believe, right?

And by 2001, the only people who hung around Silicon Valley were the real believers, right? The people who wanted to grind, who knew that this was a decade-long journey to build something that was durable and important and that made the world a better place. And it was a pleasure when the tourists left and the people who really believe stayed, right? And that period of investing, 2001 to 2005, sure, we had a lot of challenges. Interest rates were higher, we had a recession we lived through. But I will tell you, we all knew that the Internet, those of us who stayed knew that the Internet was going to be as big as we thought it was going to be. And now it was just about much more sober investing to get you there.

By 2007, of course, like humans, it takes about ten years for us to forget the sins of our prior period. 2007, we saw some silliness start reemerging. All of a sudden, public market investors thought this venture thing was easy and they started throwing big checks around. In 2007, I was on the board of Zillow, and I remember when this happened at Zillow. But then 2008 hit and again, it wrung everybody out of the system, because if you don't have deep conviction at a moment of peril and distress, you are gone. Because the only people who have the stomach to stick around are the people who really believed in the first place.

And so what came out of 2008? I launched, there are always a lot of distress-era funds, by the way, Andreessen started at the exact same time that we started Altimeter. I remember being at the Allen and Company conference with him in Arizona, and we're both passing around our decks to raise money. Andreessen and Altimeter, and by the way, he's an incredible human and built an amazing business. And to even be in the same conversation, I feel inspired by.

But then we all lost our minds again, right? And this started to build from 2009 to 2013. 2014 was really again, the golden age. We had these two super cycles taking off. We had these mobile devices, we had cloud taking off, but we still had the post-traumatic stress of 2008 that kept pricing in check. So like, risk reward was fairly distributed.

But by 2015, 2016, markets going up, everybody's like, this is easy. Again, all the tourists show back up. And man, I thought by 2000, Bill Gurley thought by 18,19, man, this thing is so overcooked and we hadn't even gotten started. I mean, then we had this thing in March of 2020, we have COVID really hit. We think it's the end of the world. But no, it wasn't the end of the world. The fed went all in. Congress went all in. We jacked ourselves up on this super Red Bull high, right? Interest rates go to zero, and the only assets people wanted to own were risk assets. So the more unprofitable you were, but faster growing, the higher the multiple, right?

And those of us who had been through that period in 99, 2000, 2008, 2009 we're like, this does not make sense. This is not sustainable. The market is being manipulated by the Fed. And by the way, I think for good reason. If you remember, in March of 2020, we didn't know if COVID was ebola. We didn't know if it was going to wipe out humanity. And so what they were worried about is that we all just stop. We stay home, the economy literally stops, right? And they had to force everybody to take risk. They had to force everybody into the equity pool, if you will, to keep the economy going.

So at the start, I think totally justified. By the middle of 2021, I think the Fed lost their mind, stayed the course way too long, caused hyperinflation that we then had to come back and battle. But we knew this was upside down on the All in Pod. I was talking about this a lot throughout 2021, by the end of 21.

And then we had the reset in 22, a tech recession, massive reset in pricing. But here's the interesting thing, because I had to set all that up just to answer the simple question like, where are we now? We never got back to the level of despondency that we had in 2001 and 2009. I mean, there, it was washed out. People were so beaten up. They had looked into the abyss, they had witnessed death, and they wanted nothing to do with it. It was a scary feeling, right? I think 2022, yeah, people started to say, oh, we need to tighten our belts a little bit.

But I would say we only ratcheted back to what I would call maybe 2013, 2014 levels of fear and pricing. And now we're entering another super cycle in AI that obviously has some dynamics unto itself that we'll get into. So listen, I think we formed a lot of bad habits during low interest rate environment, and during ZIRP, I think it led to excess.

You know, I wrote this letter to Meta, time to get fit. And I said to Mark at the time, this is an open letter to all of Silicon Valley, right? And I encourage you to read the letter. And then I encourage you to read his response letter that he wrote in March called Year of Efficiency, where he said, flatter is faster, leaner is better. And man just his leadership, Elon's leadership, and others in demonstrating the need to wring out the excess out of these businesses and all the people that they were hoovering up.

That made it impossible to build a startup because the talent was getting overpaid in all these places. And so why would you ever leave? Made it more difficult for the entire ecosystem. And now that whole cycle is reversing and I see the right heartbeat coming back. But listen.

Set a clock, set a timer in your calendar for seven or eight years from now, because we'll forget all of this again and we'll all lose our minds again. And there'll be some event in the world and we'll all get jacked up on Red Bull again and make sure that you say, god, I remember that conversation about 2022 because that is the time you need to be dialing it back.

Elad Gil

That's when you sell and retire. I think it's kind of interesting because if I look at private markets versus public, because public markets have adjusted by 50% to 80%, depending on the company, private markets haven't really shifted much. A, because they're illiquid, and B, because people raise so much money.

And so I think one dynamic that we haven't quite seen is any real carnage happen in private tech, and so we can come back to that in a second. And then secondly, to your point on AI, I feel like the tech economy was almost like an oil economy and we ran out of oil and we discovered shale. It's sort of the thing that kept everybody going and said okay, now we still have energy. It sort of reinflated the entire ecosystem and environment because it's a fundamental technology shift.

Do you think the next year or two are going to see pain in private markets?

What do you think is going to happen to all the companies that raised a couple of years ago that may not have strong business models?

Do you think any reckoning has happened yet? What sort of reckoning do you think will come, if any?

Brad Gerstner

Well, I think we really have to bucket this. We have to cohort it because there's a lot of stuff in there. First, let's talk about all those companies that raised Series A's at inflated prices over the course of the last three years. You know, they're all in market, and unless they have AI in their name, there's no chance they're getting an up-round done, right?

All of those companies, if they raised at a billion dollars with no revenue, if they don't really have product market fit, they probably just disappear. Because the reset that's got to occur in those businesses is so demoralizing to the people in those businesses, so demoralizing to the investors. It's almost better to just shut it down, and start over.

And if you look at the PitchBook data that I think JCal and I tweeted last week, but I think they talked about on the pod, I think this most recent quarter or month, we had 239 shutdowns. So if you look at the rate of shutdowns at Silicon Valley, it's going pretty parabolic. And these are the early-stage companies that were pre-product market fit, that were raising at Series B, Series C, and Series D valuations. So the best case scenario in those companies is you're going to do a big, and again leave the AI kind of deliciousness aside for the moment, I'm talking about the non-AI businesses. Those will all do big down rounds. The best of them may be a 50% down round if they want to raise new capital. Unfortunately, a lot of them, I don't like this pattern, but they're extending converts.

Listen, a convert is very dangerous. And the reason the convert is dangerous is that it comes home to roost at some point in time. And just expanding your convert, you're just denying reality. And delusion is not a strategy, right? So reprice the business, get the price right, and if that is not satisfying to you, if you don't own enough of the business at that point in time or whatever, call it quits and restart, right? Because you got to have the foundation right in these businesses. Otherwise, it's very difficult. Okay, so that was the early stage, guys.

Elad Gil

Over what period do you think that plays out, by the way?

Do you think that's the next year, the next two years, the next three years?

Brad Gerstner

I think I said last year it was a three-year journey, so now I think it's a two-year journey, right? It's when these companies run out of money and need to come back to market. And I see a ton of them in market. And just to put in perspective, for a firm like ours, our pacing over the last 24 months has been one 10th to one 20th, our normal pacing, because, listen, I can't do this. This thing in the public market corrected by 80%. And now you're coming to me and acting as though interest rates had not reset and that nothing happened in the public market.

I can't live that delusion. So until as buyers and sellers, we have an honest conversation, and by the way, I'm not looking to steal anything because I've been in your shoes, but we have to have an equitable distribution of risk and reward, right? Because I have a fiduciary obligation to the people who give me their money. I care about these institutions. They’re colleges, they’re foundations, they’re hospitals, they’re state pensions, teachers and whatnot. I'm going to fight like hell for these people. And so I can't just pretend nothing happened. So the early stage, I think, is going to have to weather this. And I expect the PitchBook data on shutdowns will continue to climb. By the way, it just started going up over the course of the last few months.

To your point, we're kind of just getting started now. Let's talk about the thousand unicorns. So these were the companies that were valued between a billion dollars and $50 billion that were further in their journey. Most of them had product market fit, had real revenues, maybe 50, maybe 100, maybe two or $300 million in revenue. But they were valued at 50, 70, 100 times ARR.

So I'm going to use software as kind of our analog here. Follow Jamin Ball at AltCap on Twitter. Jamin does probably the best analysis of software. He's a partner of Altimeter and his Substack is called Clouded Judgment. And it's just training in the art of thinking about this. And he has 1 foot in venture and 1 foot kind of in public markets.

But you'll see that for high-growth companies, profitable companies in the public markets, their revenue multiple is now about eight and a half times. Okay, so now just run the math. If you're raising money at 3 billion, 4 billion, 5 billion at 70 times, just run the forecast out. How long do you have to sustain that growth to grow into the public market multiple? That ultimately is your exit. And you just realize very quickly it's impossible. No software company has ever done that, right?

Worse yet, now we're entering an economic slowdown. And so these high-growth companies are now slowing down. So they have these multiples up here and they're now slowing down. So they're coming to you. And they're no longer growing at 70, 80, 90%. They're growing at 20 and 30 and 40%. You say, hold on a second here. Amazon AWS, which is on a $70 billion ARR run rate, is reaccelerating from 12% to 20%, right? You've got these massive businesses like a Snowflake, still growing at 35 or 40%. So you have 100 million in revenue. You're only growing at 20%. You're not profitable, and you want a premium multiple to these players. No chance.

So, again, I think that entire cohort, if you just wanted a rule of thumb, again, not any specific company, but you take Instacart, an amazing business built by amazing friends and people who killed it. But their last round in the private markets was $39 billion. Today in the public markets, they're worth six and a half. Why should any of these other companies have a better situation than that? That should be your case study. That should be your benchmark. If you're one of these unicorns and if you want a higher multiple or less adjustment down than they had, then show us why you've outperformed those expectations.

So that leaves us with only one thing, and that's a slice of the market called AI. And I love the shale analogy. So I think that what makes AI so difficult from a venture capital perspective today really is two things. Number one, we have this massive reset going on around interest rates, normalization of multiples, these drawdowns. But at the same time, we have excitement that only comes to us once a decade, that the Internet brought us, that cloud brought us, that mobile brought us, and we have it. And I think we're right that AI is that big.

Augmented intelligence is going to be what we probably invest against for the next two decades of our life. And so you can have that level of excitement just like we had about the Internet. 97, 98, 99. And you could be deadly wrong from an investing perspective. Think about this. I've used this analogy. 1998. Lots of us thought the Internet is going to be huge. And one of the number one winners or areas in which you can win will be search.

Okay, why? Because they're organizing the world's information. It's the point of entry to the Internet. You put up a little toll booth, you collect a little toll, whether it's per transaction or whether it's subscription like, great place to be. So gather all these people, get them on your platform, you be the toll booth. At the time, it was AOL and Yahoo that were really collecting the tolls, but we all knew there were a lot of challenger technologies, Alta Vista, Lycos, Infoseek, Go, AskJeeves, Google. And so every venture capital firm wanted a logo.

Okay? So what'd you have, you got those two things right, and they all had revenue. And all their revenue was going like this. Okay, why? Because it was easy to get revenue at that point in time because advertisers all wanted to be on the Internet, and these were the only games in town. So all of a sudden now for every venture capitalist, you had all three of those boxes checked. You were like, count me in. I have to have a logo, because otherwise how can I go back and talk to my LPs? And they say, well, who are you investing in search? And you're like, I don't have a search logo. They'd be like, you're an idiot. You're fired. So everybody thought they had to have a search logo.

The truth of the matter is, all those companies, you got all of that right, and all of those companies still went to zero. And you could have waited until 2004 and bought the Google IPO and captured 95% of all the profits ever generated by Internet search, okay? Now, that's one of the reasons I like to do both venture and public markets, because I did buy the Google IPO, right? I was like, man, this is way easier than that venture stuff. Just ride this pony for the next decade, which we did.

But the truth is, I kind of have this deja vu right now. Like, when I look at you and I have talked about every foundation model player in the world, and some people might say, oh, Brad, you're grumpy, you're over the hill. All you talk about is valuations and multiples. But no, I believe these are going to be really instrumental in changing the game. The only reason we have this level of excitement is because of the convergence of all this data set free, silicon that's come together, supercompute, and these transformational models that allow us now really to extract meaning out of an unlimited amount of data, right?

So trying to sort through that and say, okay, if that's true, what is going to have durable value capture? Am I getting head faked? Is this Alta Vista? Is this Lycos? Is this AskJeeves? Or is this Google? Right? And why did Google ultimately capture all the value? Because everybody went to Google, right? They owned the top of the funnel. They became the place. Not only did they own search, they were so crafty. They said, you know, this address bar that people used to actually type in URLs, we'll just turn that into a search box so that anything you type in there becomes a search too. I mean, the level of execution to build that natural monopoly, right? And they captured over a trillion dollars. I would argue it's over 110% of Google's value today is their search business.

And for the first time in 20 years, we have a challenge to that. So I was talking at the Javits Center in New York last week, thousand people with Jason and the audience, and we said, how many people have used ChatGPT in the last two days? I'll ask this room, how many in here have used ChatGPT or Claude or some equivalent in the last couple of days? Everybody's raising their hand. How many people used it as a replacement to what you might otherwise use Google for or search for? Okay, and about half of the people raised their hands, which is about what I would expect. Okay? And so I can tell you, for 20 years, if I asked that question, no hands would go up.

There was no alternative. Right? And so all I'm saying is, for the first time in 20 years, there's the possibility that something else may emerge as the top of the funnel for how we live our lives, how we plan our travel, how we get answers to questions that we have.

So I'm sitting here with my mobile device, with my team, and we're saying, okay, here's the challenge. I want to find the coolest, hippest business hotel on the West Side of New York near the Javits Center, and I want to know if it's available tomorrow night, and what it costs. You use Claude, you use ChatGPT. You use Meta AI. You use Google. Go. You use Booking.com. Okay?

And it's very clear to me that the promise that can now be fulfilled is the promise that Rich Barton and I started with 23 years ago. Rich was this brilliant engineer at Microsoft, Bill Gates. He was working on CDROMs on the computer. He's building travel CDROMs for the computer. He goes to Bill, he says, ”Hey, Bill, CDROMs, that's not the answer. We're going to put these things on Microsoft, they had an MSN service that competed with AOL. Let's put it on there, let's turn it into the internet.” That's what I feel in this moment. I said Rich for the first time in 23. Remember you wanted to put the world's smartest travel agent in everybody's pocket and instead all we got were ten blue links. Well, now we can fucking do it. Like, we can put the world's smartest travel agent in everybody's pocket, where we interact, we ask questions, right? They come to understand me. They know my calendar, my travel schedule, and guess what? They can book it, right? Because action bots are like months away, not years away. They're going to close the loop for us. So this gets me really excited because those are multi-trillion-dollar opportunities that are sitting out there waiting to be disrupted. And I think, again, we can go through a bunch of other things. So I would say valuations there are spicy and they're spicy because those of us who are looking at this saying, holy shit, this is shale, right? This is a revolution in the discovery of oil, this changes everything. I think there's a good chance that we're right. But I also am haunted by the AltaVistas and AskJeeves of the world. And I think being too early in too many things can be tantamount to being wrong.

Elad Gil

Yeah, I think directionally it's kind of clear that this is a big wave. And so I think everybody who's starting a company or investing in companies is thinking the thing they're doing is the right one. But the reality is, if you miss Google, then you missed most of it, right?

And so it's back to like, what are the characteristics that suggest that something is the right company? So you mentioned durability is an important thing. So say that you come in and you're about to lead a series B or C in a key AI company. What are the characteristics you look for? What do you view as those metrics of durability? This is actually Google and not Lycos, right?

Brad Gerstner

I mean, these things are some just enduring attributes. First, business is one of the hardest competitive sports in the world. So if you're not dealing with maniacs who are willing to sacrifice everything in life to win, then you probably, no matter how good your product is, no matter how good your idea is, no matter how good your head start is, you probably don't win. So at that stage, I have to know, what is the wedge that makes this group of humans specially entitled to win in this thing?

Then, of course, secondly, for Altimeter in particular, one of the things that we spend a lot of time thinking about is I think it's probably equally I don't know if you have parents who started a restaurant, but I always look at somebody who starts a restaurant. I'm like, that's one of the hardest things in the world to start. Emotionally draining work 24x7, kill yourself. Barrier to entry is low. It's probably as emotionally and intellectually challenging to start a restaurant as it is to start Google, right?

Is what we're betting on worth the risk and the grind and the probabilities associated with success? So I say if we assume everything you're saying is true, how big is the prize? Are we talking about a trillion-dollar prize, all of search? Or are we talking about a small prize? Right? And then obviously that factors how we think about the price of entry. Price of entry matters. It's got to be a fair distribution of risk and reward.

And then I would say another hallmark of Altimeter: we talk a lot about our culture of essentialism, the art of doing less better. And I encourage you all to read the book by Greg McKeown. It's a great book, but to me, there are different strategies and find your lane in venture.

I know people who place a lot of bets, right? And then they just see, they're like, okay, I'll pour water on the ones that are sprouting. We probably spend more time as anthropologists, researchers. We wait, we study, and when we see the real indicators emerge, then we're pretty aggressive in terms of really making that an important bet within the portfolio. So we have less ideas, but we have more wood behind each swing of the bat, I think, within our portfolios.

So those are some of the attributes, I think, that we think about. Are they killers? Do they have a great idea that if all the things are true, it's going to be a really big outcome? And then finally, are these the people that we want to be in business with for the next ten years? Right, integrity. Do they think there should be balance between a fair and equitable distribution of risks? Are they going to have our backs? Do we want to have their backs? Because this is a long journey, these are hard journeys, and life is short, so be in business with people that you actually want to spend time with.

Elad Gil

I think related to that, if you see the shift that's coming through AI societally, there's going to be potentially a lot of changes in terms of how we think about work and how we think about equity and sort of the nature of the economy and other things are going to shift over time.

And I think one thing that you've talked about a lot is really how technology and people working in technology should take a responsible lens on innovation, how they should engage with society, how they should try and push things forward. There's so much pushing technology and optimism forward.

How do we also make sure that society benefits holistically in that?

And one thing that you talked about recently that I thought was really interesting and inspiring was InvestAmerica, which I think is an exciting program in that direction. Can you talk a little about that? I think it kind of ties into some of these things.

Brad Gerstner

You know, first we're all super privileged, right? We're in this room, and we get to play this sport in one of those most magical times and magical places in the history of the world, not just the history of capitalism, like the history of the world. Picture this. We have an hourglass. Each grain of sand represents 10 million people. The bottom of the hourglass is nearly full. 110,000,000,000 people have lived on this planet and are now in the bottom of the hourglass. And for all of them or for 99.8% of them, they never saw a single invention during their entire life. Their life cycle was shorter than the invention cycle. And now we have a few grains of sand going through the hourglass every year, and we have only a few in the top, just 7 billion in the top. And we get to live at that moment in time where we're inundated daily with new innovations.

And if you look at almost all the advancements of humans, literacy, longevity, quality of life, right? Average GDP on the entire planet, like, this is the best time in the history of the world to occupy the planet, and we're the ones who get to experiment, to invent the shit that moves it forward even more. And AI is unlocking that at an accelerating rate. All that is true and makes me a huge optimist, and that's why I'm passionate about what we do.

I think what we all do matters, but I'm also eyes wide open to the fact that what we do does not impact everybody equally, right? In fact, some of the things we do adversely affect certain people's human progress, by definition. So work with me on this, right? The Industrial Revolution. If you were in the business of making horse-drawn carriages, or you were a blacksmith repairing the horseshoes for horses, you were out of business. Now scale that by a million times. Think about the people who are going to be displaced by AI, right? Like what the Industrial Revolution did to craftspeople, we're going to do to white-collar employment with AI call center engineers getting 30% to 50% more productive. That's going to change the rate at which now there will be other new and great jobs.

I'm with Andreessen, and I think I'm in the optimist camp on this. But those dislocations happen fast, and then society rebuilds and backfills. So part of my thing, particularly for this country, because we're in the privileged position to do this, I think that everybody should share in the upside of this grand bargain that we're all working on together. And so it's a very simple idea. We have 3.7 million children born in this country every year. The federal government should set up a seed investment account for every one of those children. Seed it with $1,000 in the S&P 500.

Because today 70% of families and kids will never benefit from savings or investments. Warren Buffett calls compounding the 8th wonder of the world. You want to know why wealth gaps exist? Because the fact of the matter is, if you're a winner today, Mark Zuckerberg, he sells to 3 billion people. Rockefeller, the prior age of robber barons, had 20 million customers at most. So just think about the scale at which we're now operating. Think about the people who work in call centers or work in places that are going to be displaced by AI.

So we bring everybody under the tent, right? I think this is going to be a golden age for this country, a golden age for our markets. This is less than 1/100 of 1% of our national budget to give everybody skin in the game in which we're all playing. And if we can afford to spend another 60 billion on Ukraine, or 3 billion for 20 countries around the world every year in foreign aid, think about this. The 60 billion going to Ukraine. And that's a totally separate debate, whether you think that's worthy. But that would pay for 50 million children in this country, almost every child for the next 15 years to have a starting account, a seed capital account, right.

And if you grow that and compound that over time, if you start with $1000 and you add $750 every year. So I've talked to Dara Khosrowshahi at Uber. He's like, well, we'd probably match that for all the kids of our employees. Right? I was talking with Michael Dell. Yeah, that's an interesting thing. We would consider that. So now we have a match and parents can start saving for their kids. Then by the time you're ten years old, you have $15,000 in that account. But by the time you're 30 years old, you have $150,000 in that account. By the time you're 50, you have a million dollars in that account.

We have a crisis in this country that less than half of people under the age of 40 believe in capitalism. Okay? Pew and Gallup are showing polls that say just in the last three years, 10% more people lost their confidence in capitalism. We have to solve that crisis. Ray Dalio talks about the rise and fall of nations in this moment, but it's not static. There's a dynamic system that we can actually do something about.

And so we have a broad coalition coming together, a bipartisan coalition coming together to support the legislation. We've got the funding. Follow us. I brought a little take home I'll leave you with but follow it at InvestAmerica24 on Twitter and look for the ads that will start hitting this political season, and legislation to introduce next year.

And yes, I think to answer your question, we are privileged to do what we do. We have a voice. We have reach. I think we need to be part of these solutions, whether it's AI regulation, whether it's issues of national security, whether it's issues if we look at the regional banking crisis that we had earlier in this year. I think we do. And we should have voice, and we should use that to bring everybody along for the benefits that we're seeing.

Elad Gil

That's great. That's a very inspirational place to stop. So thanks so much for joining today.

Brad Gerstner

Thank you. Happy to do some Q&A. Thank you.

Audience Speaker

So many questions. That was amazing. Thank you. I guess one, to sort of where you ended with America's place in the world. I think it sort of has potential to be a really special period for capitalism. But we also have this problem you mentioned about people not having faith in capitalism and people not having faith in America. And we have this kind of ballooning debt problem, two wars on two fronts happening. You can see how these things might overstretch America economically, and most importantly, the faith in the US.

And so how do you think that plays out? Where do you think AI plays into it?

In theory, winning the AI deflection makes America sort of gain its place in the world. I just would love to hear your thoughts on that.

Brad Gerstner

I would say first, and you probably heard us debate this a little bit on the All in Pod, where I'm affectionately the fifth bestie, which means I'm not good enough to be in the top four, but if somebody goes down, they pull me off the bench. What I would say is this. I think having traveled the world and looked at entrepreneurial ecosystems in China, in the Middle East, in Europe, etc. I think one of the things that's deeply underappreciated about this country is the genetic imprint. This country was founded by entrepreneurs. It was founded by risk-takers. The move west was risk-takers.

Every step along the way was basically an entrepreneurial journey. We set up our entire system. Even if you think about the system of bankruptcy as an incentive system to be an entrepreneur that even if you try and fail, we value you so much that you don't have to ruin the rest of your life. Thomas Jefferson declared bankruptcy four or five times, right? Like, that was a deal we made in this economic system.

So first, I think that exists. I think the pulse is as strong as ever. With that said, I don't think there's a natural entitlement forever. I think you need to recognize this specialness and you need to protect it. I look at it and I say, you've probably heard us talk on the pod. When you think about the dollar's reserve currency, when you think about interest rates, these are all relative questions. Relative to who else? The yuan. Right? I was just in the Gulf countries where they've been pegged since the early eighties to the US dollar. And I asked the heads of the Fed in Bahrain, in Saudi. I hear a lot about the yuan. Would you consider repegging? They're like, are you kidding us? Right? Like, the yuan is manipulated. The US has got the most diversified strong economy in the world. The future of technology is happening in the United States.

At the end of the day, why do you invest in a company? Because you're like, fundamentally they have the best developers, they're building the best products. They may have a bad quarter, but it's an engine of innovation that will continue to grow. And I think that is the perception I have of this country. And again, there are things we need to do better. We can't afford $2 trillion annual deficits. We have a $34 trillion national debt. Right? Like, we need to be serious about addressing these things, but we can. We will, and we should.

When I think about things like InvestAmerica, you also can't be against everything we do, spend a lot of money. But we have a choice. Do we need to spend $5 billion on the next weapon system? Or how about seeding every child in America to bring them into this system of capitalism? And so I think we need a fresh look, right, top to bottom, at where our spending is. And it's going to mean, listen, prior generations borrowed from you guys and borrowed from your kids. That debt is going to have to be paid, right?

And so the good news is we are among the few countries on the planet that have a free cash flow coming out of this innovation super cycle. We're the largest producer. We're now a net exporter of oil, like we have a lot of natural gifts, right? And the intellectual and ingenuity is the greatest one. So I'm optimistic about that. But we can't turn a blind eye. We can't take it for granted. I think the state of California has done too much about that. They need to get it back on the right track. And I think that listen, the last ten years, I said, time to get fit to Mark Zuckerberg. I said it to everybody in Silicon Valley. I can write the exact same letter to the United States federal government. It's time to get fit, rein in the excess of the free money period. Right? And this is just basic math. We can do this.

Elad Gil

I think if you look at our world and data too, there are two things that stand out. One is what Brad said earlier, where if you look at almost every aspect of humanity in terms of global poverty, child education, education of girls, death by disease, famine, etc, we're basically living through the best period of time that humanity has ever seen.

And then if you ask people in different countries how optimistic they are about the future, the poorer and worse off the country is, the more optimistic the people are. And when you get to the west western Europe, the US, etc. They're the most pessimistic about the future.

And you could argue that there are two reasons for that. One is viewpoints around participation and capitalism or other things. But I think a more important one is really good times tend to distract people from what really matters, and things get too easy and you take your hands off the steering wheel. And I would argue the last decade we took our hands off the steering wheel in all sorts of ways, the fiscal responsibility side, but also how we think about the roles of universities, how we think about the world philosophically. And so I think there's going to potentially be a hard reset because you start squandering all the positives over time.

Brad Gerstner

You can have hard resets or you can grow into it. And I think if we do this responsibly, we can grow into it. Remind you it was only 1999 that we had a surplus in this country, right? And I was with Bill Clinton this summer, and I said, you know, how were you able to govern in those moments? And he's like, because I was the governor in the state of Arkansas, and we had people who lived up in the mountains. He's like, I had to go into rural places and talk to people who had guns on the counter, right? And I couldn't ignore those people. I couldn't be an elitist. Like, I had to bring everybody together.

And what this country needs, it's so evenly divided, we need people who can bring people together around. Pragmatic solutions. InvestAmerica. I've started now four companies. I've been in politics for a long time. It has one of the highest product market fits I've ever found. Republicans and Democrats all listen for a minute and they're like, why has this not happened? People on the Hill, why has this not happened? Let's do this, right? And so that's just a pragmatic solution.

I went there with my son this summer, my son Lincoln, and we met with Speaker McCarthy. And as we walked out of an hour meeting talking about how to make this happen, I turned to him, he's 15 years old, and I said, every single thing that's ever happened in this country started with a conversation like that. A concerned citizen, a concerned community member, they started the conversation just like every company in Silicon Valley started with one person pulling out a napkin. Lots of people may have had the idea, but one person started. Right. And those are how movements are started.

That's why I think it's so important that we not only build the engines that drive the country forward, but we bring that same passion and intellect to the much bigger issues that the country faces.

Elad Gil

I think I just got convinced to run Brad for Governor.

Brad Gerstner

No, I'm not doing that.

Audience Speaker

Thank you so much. This has been awesome. Longtime fan and student of Altimeter.

I have eight questions that I'm going to pick one from, and I'll preview the second. The second takes on OpenAI 30 billion and 90 billion. What I'm actually asking you is your latest 2023 framework for dealing with uncertainty. You've said if it's not a fast yes, it's a quick no. And then I’ve also read Essentials twice on your recommendation. Thank you. And I'm just curious what your 2023 framework for uncertainty is, given you're investing in AI where there are literal funding risks. Can we get to GPT5? Like the Oppenheimer situation. Then you also invested in Grab. You took Grab public and 25% of their GMV was, I think, in Indonesia. So I'd love to hear your 2023, now we're a year or two in AI, what's your latest framework for uncertainty as a top investor on the planet?

Brad Gerstner

Well, I think that it's a rare breed that can be both a venture capitalist and a public market investor. When you think about venture capital, in many ways, it's the art of the possible. And when you think about the public markets, it's the art of the probable. Right. But we often just look through the lens of why the All in Pod grew out of poker. And why am I heading there tonight? And do I look forward so much to our Thursday night poker games? I mean, everything in life is a distribution of future unknown probabilities. And that's fun, right? And the same framework applies in venture that apply in public markets, that applies in poker.

Right. And we're continuous learners. And how can I, as an analyst, gather more and more? How can I make this single processor have more data, and run faster cycle times to make better decisions? As I sit here at the end of 23, I'm just sobered and reminded that ultimately, all valuation frameworks end up in the exact same place. It's a reasonable multiple relative to the risk-free rate of return of future expected cash flows. That's it.

Revenue multiples are total bullshit unless they result in future free cash flow that you're applying a multiple to that compared to the risk-free rate of return, providing you compensation for the equity risk that you're taking. Right? And so I would say I know people who, after moments of great distress, get paralyzed by fear. So it really helps to be an optimist. Buffett's like I've been an optimist for 50 years, and it served me well. I agree. It's really important. If you get up every morning pissed off about the world and think that tomorrow is going to be a worse place, you're going to be a terrible venture capitalist, and you're not going to be a very good public market investor. But I will hire you to be on my short team, to look at good short ideas, because that is just constitutionally something that is different.

But at the end of the day, Warren Buffett, one of my heroes, guys like Paul Reeder, value investors, guys like Seth Klarman, I'm an optimist, but I'm also kind of a value investor. Like, I need to understand why the distribution of probabilities is fair. So in the public markets, last year you could have bought Facebook, which is one of the greatest businesses in the history of the world, trading at six times EBITDA, right, at nine times fully taxed earnings. Mark gave you the blueprint on December of 22. He said we're going to get fit. You knew how much they could save. You knew how much margins could expand. You knew how much he was spending in the Metaverse. So that is a puzzle I love putting together and I love the conversations I get to have with some of these incredible founders who are still running some of the biggest businesses on the planet.

But we've actually, relative to our peers, we've done very little in AI so far this year. And I have to tell you, it's very hard because I'm as enthusiastic about all these companies and founders at this moment in time as anybody. But at the same time, when I look at some of these opportunities, let's just take OpenAI, because I think it has the best team in the business. GPT4 is clearly the best if not one of two of the best models that exist out there today. But if you're asking me to invest at $85 billion to buy common stock, that's capped on the upside, right?

Now we have a deal. It's like a poker hand. And I can say, okay, I know what my upside max potential is. Right? I can then plot a distribution of likely outcomes. How many companies have ever in the history of Silicon Valley been worth durably more than $100 billion? What do I have to believe to be true?

Okay, so they have two businesses, an enterprise business and a consumer business. Well, in the enterprise business, they have to go head-to-head with Microsoft, right. They have to go head-to-head with Google, head-to-head with AWS. So we went out and we talked to 200 customers and just like, what are the customers telling us? Right? And the customers are telling us we can't send our data to OpenAI. Right? Like, a public company can't do that. So we're going to work with Azure OpenAI, which, by our accounts, benefits Microsoft, and doesn't benefit OpenAI? Or we're going to stay at Amazon or we're going to work with Snowflake.

We have to apply some discount rate to our model. About how big are they likely to be in enterprise, how hard is it to challenge the biggest incumbents? You got to go build a sales force like who's building it? On the consumer side, which is where I get a lot more excited. about OpenAI, is I say you are the verb. I would rename the company ChatGPT. I would say we're not going to spend a minute thinking about enterprise, we're going to recut our deal with Microsoft and just get some of the goodies that come out of enterprise. And I would go all in on building the world's best fiduciary agent to serve all of us in a way better than ten blue links, right? Because there's a trillion-dollar prize if you can displace Google at the top of the funnel.

But it is going to be a heroic battle. And Google's not asleep at the switch and Meta is not asleep at the switch. And you got new incumbents like Inflection, you got folks like TikTok. You know, Bing is building a really good agent. So there are going to be a lot of people in this game, right? And if you want to win that game, which is the biggest prize in all of technology, you better be singularly focused with the best team in the world and building the best model.

So, again, at that, I like all these guys, they're friends, I have deep respect and it's really hard to say no, but at the end of the day, I'm building a portfolio to manage investment returns, right? And we've compounded at some of the highest rates, I think, on the public markets and in the venture markets now for 15 years. And I didn't get there by just investing in things that I thought would be sexy to have in my portfolio or logos that I wanted to collect. I have to see math that ultimately works.

And I reserve the right to change my mind on OpenAI if we learn additional facts or if the price was less, if the conversation today was 10 billion or 20 billion, if I didn't have uncapped or I didn't have a capped upside, if I wasn't buying a common security. I mean, there are a lot of things that you could change about that equation that would absolutely send me running in the direction of wanting to invest in them.

And so you can go through everything in the AI stack and I would say I wish we had put more money into the ground. It's been difficult to find things that I thought had an equitable distribution of risk and reward in AI. But I do think coming out of the summer we're starting to see a bit of a reset in AI because ultimately these things are like if you're a software-oriented company in AI, you ultimately are a software company with software like economics that has to sell into the public markets. And we know what those multiples are. If you're a consumer-facing company, then we've invested in lots of consumer companies over the years, from Google to Booking to Zillow to Bytedance/TikTok to Facebook. We know the way that works, so we can just pencil it out.

And so my suspicion is that some of the lather will go away over the course of the next year or two. But in the short run, we're probably overestimating it. And we're in a bit of a little bit of hypishness around prices, but over the medium to long term, we're probably underestimating it. The winners, trust me, the winners of this will generate a lot more than $100 billion of enterprise value from the prize that they're going to win. There are trillion dollars in prizes to be won by the winners in this and so earlier stage, if you're investing like this guy's lucky enough to in the earliest stage, those are a different type of bet, different size checks.

But as you scale up and the price scales up, you have to reorient. So an angel check, where I'm writing 50 or $500,000 into a company, the only thing I'm underwriting there are these killers. And it's a generally good idea. And so you could place a bunch of those bets on the board. But if you're writing a 100 million or a 300 million dollar check into a company, then you have to know that there's, I don't even call them venture companies anymore at that point, I call them quasi public because they get shopped around to a bunch of people, right? We all have big stacks, we can all write the checks. And so at that point in time, there's much less arbitrage in the market, right? The market becomes way more efficient. And now I have to compete. Just like in the public market, it only takes one person to price that last security. And so I don't know who that person is. I have to price it myself and say, are we set up asymmetrically to win?

And so hold these two truths that are intentioned just like they were in 98, 99. This will be bigger. It will have more impact on humanity than probably everything that came before. But it doesn't mean you have to own it all now. This will play out over decades, not days. And so that's the way we're thinking about it. It's been a lot of fun and I envy the stage at which you guys all are. I'd pay a lot of money to go back and play 20 years over again. But follow the InvestAmerica thing. Get in, let your voice be heard. And thanks for having me.

Elad Gil

Thanks so much.

MY BOOK
You can order the High Growth Handbook here. Or read it online for free.

OTHER POSTS

Firesides & Podcasts

Markets:

Startup life

A brief guide to startup pivots

4 Ways Startups Fail

Founder Investors and Scout Programs

Co-Founders

Raising Money

Capital Efficient Businesses

Elad Gil — Mon, 04 Dec 2023 22:25:10 GMT

For the last 5 decades[1], every wave of technology-driven companies has had amongst them highly capital efficient businesses. The capital efficiency tends to reflect that (a) people really want your product and will pay you for it and (b) the founders are cost-conscious and frugal, and do not overhire.

Indeed, Paul Graham from YC has developed the metric of “default alive” to reflect capital efficiency as a core sort of startup metric.

Capital efficiency has existed in roughly every technology wave. Many of the largest, more important companies in the world started off highly capital efficient. Inded, capital efficiency tends to reflect an especially strong business model (and in some cases founder). Examples include:

Microsoft was bootstrapped in the 1970s and did not raise any venture capital until a round right prior to their IPO, when Bill Gates wanted a VC for his board and that came with strings attached for the VC to invest. The investment went straight into the bank account and remained untouched.
Dell was bootstrapped off cash flow in the 1980s until a similar pre-IPO round.
Yahoo and eBay famously did not touch their early venture capital funding in the 1990s - they were run so lean and profitably that they did not need to raise large sums.
Google raised a single round of traditional venture capital, before doing a pre-IPO round with Yahoo! and others.
Instagram was just a dozen or so people before being acquired by Facebook, while Zapier only ever raised a $1.3M seed round before bootstrapping from then on. These companies were founded in 2010s on.
Midjourney (founded in 2020s) is rumored to be entirely bootstrapped today.

Why are some businesses capital efficient?

In general there tend to be two drivers of capital efficiency.

Customers will pay (a lot) for the product. The “capital” side of capital efficiency is often a proxy for both product / market fit and an intense customer need. Customers are willing to pay up for a product that is important to them, and there is insufficient competition in the market to commoditize pricing or destroy the category (so the product is somehow differentiated). Pricing is often a proxy for value & differentiation of a product.
The company is run efficiently. During COVID roughly all tech startups lost their way on spending. Capital was flowing freely and teams often rapidly and dramatically over hired, boosted expenses on things non-crucial for the business, and spent wastefully. The most capital efficient businesses tend to be frugal and have a low cost approach to the world. Salaries are lower to help make equity more valuable. The founders and employees of these businesses treat the dollar spent by the business as their own money (which it is, as they are shareholders in the business). They realize that profitability gives them infinite runway and enormous freedom on decision making and future path optionality.

Frugality has felt like a lost art over the last few years - hopefully it is recovered.

When to bootstrap?

Too few silicon valley (or NY or other cluster) based technology companies bootstrap.

If you can grow organically and optimally without hiring a massive team and increasing expenses it is great to do so!
If your company is a cash versus equity business, you should bootstrap.
If your company is growing slowly and will never hit venture scale, you should bootstrap.

When to raise money?

Venture capital is typically used to either:

Build out or prototype something. This may be something inexpensive but for some reason can not be bootstrapped off of customers (e.g. a new SaaS product) or is something capital intensive that may have a giant market on the other side. The later includes things like building rockets for spaceX, or biotech drugs.
Scale something that is working. For example, you want to add sales or go-to-market functions to sell faster/better, or your consumer app is growing like crazy and you want to be able to add more compute to serve users. Uber needed to raise billions to both scale rapidly and beat out global competition.
You need the valuation for external uses. E.g. M&A or hiring (there are other ways to do this too).

In general, if you are not prototyping / proving something works, or scaling something that does work, you should not raise money.

One could argue that while too many SV/NY/cluster-based tech companies raise money, too few outside of major tech clusters do. In many cities and regions people bootstrap for too long, do not scale quickly enough, or do not think about time to winning in a big market. It is possible that non-cluster tech companies in the US end up scaling fast too infrequently.

The dangers of capital efficiency

Occasionally, you also see an SV/NY/cluster-based company that is growing really well and has turned profitable, and then forgoes building against and winning in their category. Sometimes this is the right thing for founders to do, and sometimes it reflects a lack of know-how, ambition, or aggressiveness. Sometimes, it just shows the founders had a bad experience at a company that scaled for no good reason and ruined the company culture, ability to execute, and products. The wrong lessons may be learned from bad growth and bad execution. It is so rare to actually build something that people care about, that it feels like a shame to not go win when you can - but obviously it is up to each founder and team to chose their own path. Some companies that focus on or hit profitability early then forgo winning the market - their focus shifts too much on maintaining cash flow versus growing faster to take the market.

Generically, startups are rewarded for progress per unit time, versus progress per unit dollar (all else being equal within a given burn multiple range).

Hopefully more capital efficiency returns to technology now that ZIRP and COVID policies are (roughly) a thing of the past. Many of the most important companies in the world started in a capital efficient state.

NOTES

[1] An emerging meme today is that with AI, the costs of starting a company are lower as you can augment people with AI from day 1. Past variations of this thought has occurred for many macro waves including distributed talent (“talent in India is cheaper, so your company can be too!”), cloud services (“cheaper to not build your own data center!”) and other shifts.

Notably all these shifts have impacted the costs of doing business. However, the impact tends to come with scale and the need for more people intensive operations. I am very bullish on e.g. AI applied to private equity and buyout models.

MY BOOK
You can order the High Growth Handbook here. Or read it online for free.

OTHER POSTS

Firesides & Podcasts

Markets:

Startup life

A brief guide to startup pivots

4 Ways Startups Fail

Founder Investors and Scout Programs

Co-Founders

Raising Money

Video & transcript: Apoorva Metha, founder & former CEO, Instacart

Elad Gil — Fri, 01 Dec 2023 16:46:56 GMT

I did a fireside chat with Apoorva Metha, founder & original CEO of Instacart. We discuss a wide range of topics including:

How to parse startup ideas
Founding moment for Instacart
Early things Instacart did that could not scale, but got them in business
Hypergrowth at a startup & people
Why start another company / act 2 in life?

Video:

Full transcript:

Elad Gil

Thanks so much for coming. I guess from a background perspective, I'm doing this all from memory, so undoubtedly I'll get all this wrong. But I believe you studied engineering at Waterloo.

Apoorva Mehta

That's right.

Elad Gil

And then you started a variety of different businesses. You ended up starting Instacart. And I was hoping to just hear a little bit more about the origins of the company. I know you were iterating through a variety of ideas, a variety of businesses.

How did you converge on what Instacart was and how'd you get going on that?

And I should mention one other thing. From a background perspective, you're now working on your second company.

Apoorva Mehta

That's right.

Elad Gil

And I don't know how much you can or cannot share. It'd be great to talk about that as well if that makes sense.

Apoorva Mehta

Yeah, and let me know if there are specific details that'd be more interesting. So I was a supply chain engineer at Amazon and was not fulfilled. I was like, there's got to be more to life than this. I decided that I wanted to become an entrepreneur. I moved from Seattle to San Francisco and realized that the only real way of becoming an entrepreneur is to sort of really get your hands dirty.

So I started over the course of a couple of years, I started about 20 companies, did all kinds of things. Like at this time, Zynga was growing very fast. And so I built an ad network for social games. I built a Groupon specifically for food. This time Groupon was really big as well. And all these companies failed, unfortunately.

And I was in San Francisco in my apartment. I realized that all I had in my fridge was a bottle of hot sauce. And this was a common problem for me and it didn't make sense to me. This was 2012 and we were ordering everything online except for groceries. And here it is. There's a trillion-dollar category that's still stuck offline. And so I decided that I was going to bring this category online.

I started coding the first version of Instacart and three weeks later Instacart was born. And at the time, because I was the only one in the company, I placed my order and then went to the store, picked up my groceries, and delivered them to myself. And of course, I gave myself a nice tip.

Elad Gil

The rest is history.

Apoorva Mehta

Yeah, exactly.

Elad Gil

So when Instacart got up and running, about a decade before that, there was Webvan, which was one of the most famous craters in all of Silicon Valley history. It was backed by Sequoia. It raised a billion dollars or something. It hired the CEO of Accenture and started running it and then it was like a total wipe. And it was a very asset-heavy business.

What was the reception when you told people that this is what you were going to do were people like, that's a great idea? Were they like, how could you do this? Don't you know the history? What was the reaction?

Apoorva Mehta

Yeah, I remember there was an investor meeting that I had in the early days and I walk into this meeting and I start presenting. And this is like 24-year-old Apoorva. I wanted to really stand out. The title under my logo was Webvan done right. And I started, got to the second slide, third slide, and this investor literally got up and left the room and I was like, Is the meeting over? It's very clear I didn't get the term sheet. But then he came back and he slapped this floppy disk on the desk on the table. And he was like, this has the Webvan business plan, you should go home and study it and you will never do this company. And so the reception was not great.

And this is sort of what you see with a lot of startup advice. It is very generic and you actually have to step back and reason from first principles. A lot has changed since when Webvan got started. And one of the main changes was the fact that now everyone was carrying a supercomputer that was connected to the Internet with GPS in their pockets. And so if you wanted to order your groceries, we could connect you with someone who was close to the grocery store, who could pick up and deliver the groceries to you. And we wouldn't need any trucks, we wouldn't need to hold inventory.

And so the world had actually changed pretty significantly, and yet this conventional wisdom had not changed. The reception wasn't great, but then you start to talk to people who are actually open to reasoning for its principles, and that got us our funding.

Elad Gil

It's kind of interesting because during that era there were a lot of things that had blown up ten years before that everybody thought would never work again. And then there were online versions of that. I think there was Chewy on the pet food side, but also things like payments. And so as PayPal sold to eBay, the conventional wisdom in Silicon Valley started to become that payments are too hard. And so all sorts of people avoided payments for ten years.

Apoorva Mehta

That's right.

Elad Gil

And I think it's notable that both you and Stripe got funded by Mike Moritz, who was in the middle of all this stuff back then. I think he was actually the board member on Webvan that blew up.

Apoorva Mehta

That's right.

Elad Gil

But then there was sort of this rethinking now, actually, maybe it now works in the context. So can you talk a little bit more about how you started working with Sequoia and how you met them and how that turned into a relationship, given that they were so burnt on the prior sort of version of this?

Apoorva Mehta

Yeah, at the end of the day, you want to start a category-defining company. And payments: enormous TAM. Grocery enormous TAM. And if there are companies that can be built that have a shot of transforming a category, that's a shot that you want to take.

There's this famous saying in investing, which is that most investing mistakes are failures of a mission rather than failures of commission. All right? And so early stages for a lot of companies, it really is like you're buying an option for this company to be successful.

Now, of course, as you get into a lot more details about Instacart and I'm sure with Stripe you start to understand, yes, it is challenging, but with the approaches that we've taken, it actually is tractable. And if you solve some of these problems, you can actually make it a much more successful company.

Elad Gil

What was the worst advice that you got in the early days besides just don't do this?

Or what was the best advice in either direction?

I'm just sort of curious, like, where was that outlying insight or was it just, hey, I'm just going to keep grinding and it eventually works out?

Apoorva Mehta

Yeah. I felt like a lot of startups are actually created and remembered by the rules that they actually break. If you look at a company like TikTok, for example, they did paid user acquisition to grow their user base. That was considered to not be a practical thing to grow. And so you always had to take advice and really assess that for your own company.

I think the most important advice that we got multiple times, and it took me some time to internalize, was the team that you built is the company that you build. And initially, as a founder, I was just so focused on getting things done that I wouldn't jump into any single problem myself until it was absolutely, you know, in a place where I felt comfortable with. But then you realize that you've solved that problem for that moment.

The next quarter that will break again and the next quarter will break again. And so the most scalable way to actually solve a problem is to step back and hire the team that can actually solve the problem. And that was advice that I got multiple times I ignored. But over time, that became the way as a CEO, I started to solve problems.

Elad Gil

Well, I guess, related to team, what sort of characteristics do you think are most important for very early teams or people that you hire into those teams?

And then how does that differ for later-stage companies?

Apoorva Mehta

Yeah, I mean, of course you have your core values that you hold dearly yourself that show up in the work that you do every day. And for me, for example, they are number one, intellectual honesty. Early stages. You can make a lot of wrong decisions by focusing on hype, for example, or vanity metrics. But really, if you're intellectually honest, you get to the right answer.

Number two for me is executing relentlessly, which is a combination of urgency and excellence.

And number three, are you going to be able to scale as a leader? Because in an early stage, specifically, it's a very different company quarter after quarter. And so those are the few things that I look for in early-stage companies as we're scaling.

Elad Gil

And then do you think anything that changes late? Because I think the nature of people who show up to a later-stage company is different. Part of that is risk aversion.

Part of that is now you have a brand, so people feel like it's a safe thing to go to or it's a status thing. And so how do you think about those different pools of people who both may be excellent but just very different?

Apoorva Mehta

Yeah, I think that there are a lot of people who have a strong aversion to people like that. But the reality is, I think you actually need a combination of those people because a lot of people who are in the early stage may not have the deep domain experience that is required at the later stage. And that's okay. In the early days, you kind of want the Swiss Army knives, but at later stages, those people are not able to scale as fast.

And so you do need a combination at later stages. Of course, there are a lot of companies that get stuck in a place where they're still doing the same thing, the same playbook, which obviously doesn't scale. So you do need the entrepreneurial DNA. Hopefully, you can retain that from the early stages, but it really is a combination.

Elad Gil

Instacart today is very different from where it was initially, which is usually a very positive sign for a company. And it started off as a marketplace. It's still obviously an important marketplace. And then on top of that, now you have advertising and services for grocers and other things. The marketplace side tends to be really hard to get going, and many companies struggle with a two-sided marketplace.

And in your case, you also had people who were actually doing three things in some sense, right? You had the buyers, the grocers, and then the people who are actually helping with the delivery of goods and services.

How did you go about bootstrapping something? Because it's already hard enough to do two. How did you do three, and how do you think about that?

Apoorva Mehta

Yeah, so the way we look at the business is Instacart's a four-sided marketplace today where we have customers on one side and we have retailers on the other side. We have people who are picking up and delivering the groceries on the third side. And then fourth, we have advertisers. These are the people who are buying ads. When you search for, for example, a beverage. Now, of course, once you have a marketplace, it's a virtual cycle demand begets supply, supply begets demand. It's really wonderful.

But to actually get that started, you have to solve this chicken and egg problem, which is not trivial. We were this small company based in my apartment and so getting a large Fortune 500 retailer like a Kroger or Costco to sign a partnership with us was just going to be impossible. And so what we did was we invented this thing called ninja shopping, which was effectively that we would just go to a store, buy the stuff off the shelf, go to the checkout, and deliver the groceries to the customers. And the only problem was we didn't know what any of the stores actually carried. All right, and you couldn't find it anywhere online. I was totally okay scraping stuff, but you just couldn't find it. And some of these SKUs didn't even have any online information about them at all.

So what we decided to do was we went to many different stores and picked up one of every single thing from the store, took it to a studio, photographed everything, and then uploaded all that onto Instacart. And it cost us like $50,000 to do that, to buy the entire grocery store. But that allowed us to bootstrap our supply. And so now it immediately became something that customers wanted. And I remember we did this for one of the stores and overnight our demand doubled.

And so we're like, okay, well this is clearly something that customers wanted. And so we did this for another store. Another store, another store. And we didn't know it at the time, but what we were doing by building this sort of data pipeline and collecting high-quality images and this metadata for the items was actually, this was going to become a key barrier to entry that we were building onto the Instacart platform. And it still continues to be a barrier today, and we're continuing to strengthen that today as well. So for us, that's how we got started.

Elad Gil

When you first started the company, because I feel that often when people think about companies early, they really index heavily on moats. But often it takes a while to figure out what your moat will actually be or to build it.

What did you think your moat was going to be when you started the company?

Apoorva Mehta

Yeah, I think a lot of times early-stage companies just really focus on growth. It makes sense trying to prove product market fit, you're looking at your customer retention. But I actually think that you actually have to be very deliberate about what your barriers to entry are going to be and you have to work on them just like you're working on your growth, your profitability.

For us, there were several moats that we thought about at the beginning and then over time we actually worked on those. First, doing this is actually very hard. We deliver millions and millions of unique items in any given city, just having that information is actually not something that anyone else has. Right? And to do that efficiently, while you have to navigate the people within the store efficiently. So we were like, well, this is actually technically quite complex, right? And as we continue to grow, this is going to only become more of a barrier for us, which is obviously wonderful. But then as we continued to grow, we noticed that we were becoming very quickly a decent portion of any retailer's store volume. Like in many cases, we were over 5% of their store volume. And that they started to pay attention to that.

And as I discussed, one of the things that we noticed was that as soon as we added a store, our demand doubled. So it was very clear that grocery selection really mattered. People cared about the grocery stores that we had on our platform. And so we developed very close and deep relationships with these retailers that gave us exclusivity to that supply. And so we had this technical moat, but we also built this relationship moat as well.

Then we started to build more enterprise features for these retailers. Today, many of the retailer.coms are powered by Instacart, and that meant that we were more and more deeply integrated with these retailers. All of this was something that we tracked on a quarterly basis and engulfed our teams on. So it wasn't something that we sort of just said, well, it's just going to happen. It was something that was pretty deliberate.

Elad Gil

I think you all were very thoughtful from the early days in terms of how you would develop the business. And I feel like a lot of the things that you talked about earlier are now substantiating themselves. So it's been interesting to watch that arc over time. I think many companies are less forward-thinking and you see things happen organically and sometimes that's amazing, and things organically form defensibility or other things, and then sometimes it just leaves things wide open.

I guess related to that, many companies end up with one or more moments where it feels like there's a big existential threat to the company or everything's going to fail. Did you have any moments like that?

Apoorva Mehta

Several. Yeah, I could have used fewer of those. I'll talk about one of them. I had just come back from one week off and it was so wonderful that I was able to take the time off. The day I came back, I got a call from the Whole Foods CEO at 6:00 AM in the morning. All right, at this time, Whole Foods was the largest partner of ours. They had about, I think 30% to 40% of our overall sales were coming from Whole Foods. And so of course, I was going to take the call. It didn't matter if it was 06:00 AM. And I get on a call with him. He tells me that Amazon had just paid like $18 billion to buy Whole Foods.

Now, I'm a very paranoid person, and you tend to be that when you're a founder, you have to understand where risks are in your business. But this was not in my risk bingo card and was a very short call because I didn't know what to say to him. And I was just refreshing my social feed. And as soon as that announcement happened, the next set of announcements were, oh, Instacart's dead. And we started getting text messages from investors and parents asking us if we're okay. And it kind of felt like this was going to be it. Like this was going to be my 21st failure of a company because now our largest competitor owned our largest partner.

And so at the time I called in all hands, told the team that we were in war mode and the only thing that mattered at this point was to fight this battle. In the next couple of weeks, we came together with a plan, and this was a high beta plan, very high-risk plan, which was that we were going to sign every major grocery retailer onto our platform and we were going to rapidly increase the Instacart membership so that we would be able to retain most of these customers. And me and the team, we were on the phone with effectively every retail CEO in America talking about what we could do.

We had this concept of the alliance of the willing. We were calling it internally, but really it was how do we figure out how do we work with retailers? And we looked at every single thing that we could do to get them over the line, which was how do we rapidly expand nationwide? So that we were everywhere. We were in Rockford, Illinois. We were in Lubbock, Texas. In the smallest cities, Instacart worked. And we figured out how to make the economics work in the smallest cities because that's what it would take for some of these larger retailers to sign with us. We scaled Instacart enterprise with all kinds of functionality that would make it so that these retailers felt comfortable putting their brand onto Instacart. And there was no meeting that we would not do in person. Regardless of how many red eyes we had to take, we made it happen.

And by the time Whole Foods finally left the Instacart platform, they were less than 5% of our sales. We had continued to drive a lot of growth and we had virtually every major grocery retailer on our platform. And so at the time, of course, this felt like we're not going to make it, but actually ended up being one of the best things that could have happened to the company.

Elad Gil

Thanks for sharing that story. As you built the company and as people build companies in general, they go through different types of investors. Can you talk a little bit about how that investor mix shifted over time and what you were looking for at each stage of the company?

Apoorva Mehta

Yeah, I think generally speaking, in the seed round, a, B, maybe even, sometimes C, at least for us, what we were pitching was mostly, look at the opportunity, we're going to take that over. It was mostly selling the growth, selling the opportunity, selling the dream. And it makes sense because you're talking about, relatively speaking, smaller dollar amounts.

And when you're talking about later stage, it's all about what is your next two, three-year, five-year plan and how you're performing on that. How is this company going to be valued as a public company? What are the real things that are going to change in the next couple of quarters that are actually going to allow you to hit the plan? Obviously, you're coming in with large dollar amounts. You're talking about investors who are at that point, you're sort of not looking to lose money. You're looking to have maybe a smaller on a percentage basis return, but you're looking to make sure that you're going to return.

So the types of investors that you have are different. And it actually is very helpful to have those different investors. And I think that a lot of times, as that shift happens, you really want that input from the later stage investors to see, okay, well, how is this company going to trade? What are the multiples going to be? What are the public investors going to think about the risks in the business? And how do you really address those head-on earlier in the company's journey? And I would say, in fact, it's actually very good to have that perspective in the early stages as well. And looking back, I would have definitely done that.

Elad Gil

Yeah, that makes sense. Was there any specific insight or anything else that a later-stage investor really brought to the table for you in terms of some aspect of public markets or some aspect of running the business or some aspect of dispensability that really landed?

Apoorva Mehta

Every investor, a lot of great investors have a different point of view that they bring. So what we would do is before every Instacart meeting, board meeting, we would have sort of a smaller meeting, and the board meeting is sort of perfunctory and you got to get through stuff. And sometimes there's insight, but a lot of times the best meetings are the ones right before then where you have one or two investors who come in and who are really providing you strategic guidance in terms of what if you slow down growth, what if you increase profitability? What if you think about expanding into a different category, different market and so on. And that conversation is actually the most helpful.

And for example, Kris brought in a lot of really interesting data from what they were seeing across many different similar companies domestically and internationally. And so we would have a lot of data in terms of retention, in terms of market share, in terms of conversion, cross-shopping. And we could use that to make very important decisions. And so I'd say those are the value adds that I was most grateful for.

Elad Gil

Yeah. Makes sense. Are you able to talk about your new company at all?

Apoorva Mehta

I can talk about it at a high level.

Elad Gil

Okay, sure. So, first off, I feel that starting a company is so painful. I've started two. I feel like it's putting yourself through so much in terms of range of emotion and work and like, why do it again?

Apoorva Mehta

Honestly, I just didn't want my legacy to be defined by just one company.

Elad Gil

It's a pretty great company, though. It is wonderful, it's a great legacy.

Apoorva Mehta

I'm very happy, of course, but if I'm able to do it again in a different category, I think that'd be really wonderful. And for me, that was always the motivation.

Elad Gil

How did your thinking about the early team change between the first company and the second company?

Apoorva Mehta

Pretty dramatically. Yeah, actually very dramatically. My thought was, in the early days of a company, you always want people who are actually not that senior because you want people who are in the weeds. They're actually doing the work as individual contributors. And so we hired some wonderful people at Instacart in the early days who were really some of the best people I've ever worked with. Right. But they were ICs.

And this time around, what I've done is I've brought in people who are actually much more senior, but they can scale down to be ICs. And what that has allowed me to do is actually not only benefit from being able to go into the details but their judgment as well. As well as in the event that we need to scale, I don't have to look for another manager. We already have the people who can recruit. And I feel like that is a much better way of approaching this.

Elad Gil

I think that's also a real benefit of a second-time successful founder because you can access those people. And I feel like first-time founders run into two issues. One is they can't access them, but more importantly, they don't even think about it.

Apoorva Mehta

Right.

Elad Gil

Often, if it's your first time really building something, you often see the opposite. You're like, why do I need all these experienced people?

Apoorva Mehta

That's right.

Elad Gil

They just want to talk about stuff. And in reality, they can be very valuable.

Apoorva Mehta

Yeah. I mean, you certainly want to watch out for the people who talk about stuff, and you want to watch out for those people regardless of what stage you're in the company. But that was a big shift for me. And we're certainly just able to move much faster.

Elad Gil

I guess there's the team side of it. Were there other aspects of starting this company that you thought of very differently or that were nuances or shifts relative to what you did with Instacart?

Apoorva Mehta

Well, I mean, I think as a second-time founder, you have a completely new set of challenges that you don't expect. Second-time founder, you're thinking about like, well, what if it doesn't turn out to be that big? All right, the first time you're like, it's a billion-dollar company, great.

Second time, you really want to make sure that it can meet your hurdle, whatever that might be. And another thing is that at least for me at Instacart, I didn't really care what the problem was. I would jump in now, of course, because of a lot of experience. I think a lot more about this problem is not going to go away. How do I figure out how to scale myself? And that's a pretty meaningful difference as well.

Elad Gil

Yeah, it makes sense. Well, maybe what we can do is open things up for about five or ten minutes of questions from the audience.

Audience Speaker

Something Elad mentions is the danger of spending too much time on an idea or market that's just not working. And the myth that if you spend enough time on something, it'll magically work when given enough time. And so you said you started 20 companies in two years. So obviously very fast-paced.

What convinced you each time to abandon that idea? And then what changes? What gave you the conviction with Instacart that this is the thing I now want to pursue and keep working on?

Apoorva Mehta

Yeah. That is one of the hardest things as an investor, as an entrepreneur, to figure out when your idea is not going to be the one. Because you sort of have to be emotionally invested in it to actually put in the work when everyone's sort of like, no, already is not going to work. But you believe in it. So the 20 ideas and building these 20 companies was actually very emotionally draining.

But then I came up with a process which really helped me, which was anytime you come up with an idea, the first thing you want to do is write down the top ten reasons why it's not going to work. All right? Just write it down and try to prove those things to prove that it's not going to work as fast as possible in two days, in a week, and before that, don't get emotionally invested in it. If that passes, I find the next ten reasons why it's not going to work. Now, this requires a little bit of work to actually go into the details, to figure out what are the nuanced reasons. These things take time. So you're going to invest, like, maybe a month or two.

At that point, if you have some resources, maybe it can be shorter, but it takes some time. If at that point you've gone through the 20 reasons why it's not going to work and you've proven that actually no, there's going to be something there, you then are like, well, at that point, if you're still emotionally bought in, you should really consider it. And for me, with Instacart, I unfortunately did not have this deliberate of a process at the time. But for every reason that I would come up with, I was like, well, I had the supply chain background from Amazon, so I was like, that's the hardest part in this company. All right, if I can solve that, which I knew I could. This company should exist. For me, I think Instacart was a little different.

Elad Gil

Do you think most founders err too much on the side of going on when they shouldn't or stopping too soon on ideas?

Apoorva Mehta

I think most founders spend way too much time on ideas that are really not great companies. There are most likely going to be zombie companies. And it's just sad to see that after I stepped down from Instacart, I spent about like six to nine months looking at climate to see if there's a really interesting opportunity there and looked at space and a bunch of other areas.

And these are wonderful categories of really challenging problems to be solved there. And then I would run into founders that had been working on these ideas for like, you know, in some cases more than a decade. And truly wonderful people, truly mission-driven, obviously. But if the time for the idea is not now, and so it's just not going to happen. All right. And sometimes it's hard.

Elad Gil

Yeah, I think that goes against the conventional wisdom. And I agree with you. The conventional wisdom is keep grinding no matter what, and years in you'll suddenly make it work. And usually it's the exact opposite. You just waste the best years of your sort of prime life.

Apoorva Mehta

Exactly right. It's so sad to see that at any given point in time, any moment in time, there's only a handful of really good ideas that can't actually be possible because of the way things in the world have changed. Maybe there's some sort of a disruption that's it and the best resources, the resource allocation should be focused on those ideas.

I'd say most founders should also not feel bad about working on an idea that maybe other people are working on too. All right, that's okay. Because maybe the time for the idea is right now, and maybe the way to win is better execution. I'm a big proponent of focusing on where you can make the most impact and that's sort of a handful of ideas.

Audience Speaker

Thanks for sharing all this. As you look back on all the decisions that you made, is there one or two decisions that you wish you could have gone back and changed and that would have changed the inflection of this space?

Apoorva Mehta

You know, to be honest, there's like, probably many, but I look at it as like a learning opportunity. All right? If you could have hired someone versus hired someone else would have been a very different game. Could have gone into a different category versus another one.

For me, you know, there's a lot of things we did right at Instacart, right. Sticking to focus on one category: on grocery, working with retailers as partners rather than competitors. And these decisions could have gone the other way. Right. I remember 2015 or so, Jeff Bezos came on stage and said that this sort of way of doing grocery shopping doesn't seem like it's the right solution. And the whole team was like, shit, Jeff Bezos saying this. But actually, in our model, the marketplace model was superior to our first-party fully integrated model for various reasons. And our first principles thinking was right.

Elad Gil

And I should say this was in a period of time where everybody was talking about vertically integrated companies again.

Apoorva Mehta

That's right.

Elad Gil

So you were actually being very contrarian relative to the big trend of the day, which was, oh, we'll just vertically integrate everything.

Apoorva Mehta

That's right.

Elad Gil

That was really interesting.

Apoorva Mehta

Yeah.

Audience Speaker

I'm curious how your role changed as the company evolved and what made you decide that you wanted to start something new instead of being a public company CEO.

Apoorva Mehta

Yeah, the role changed quite significantly, and it needs to be. Initially, you're obviously focused on every single problem in the company. And as you continue to grow, you need to pick your battles. You need to pick the fires that you're going to fight. Because there's just so many things going on to the point where you get to where you're primarily focused on resource allocation, making sure that you have the right people around you who are doing that, and to the point where you continue to become more sort of a figurehead.

And, I mean that not in a negative way, but you sort of need to be, as a company continues to grow, and you're obviously in the details in some areas, but you also have to be the chief cheerleader for the company. And I loved so many aspects of this. There are aspects of it that, frankly, were just not part of what I found fun. To me, coming up with a new business model that could solve a meaningful problem in the world, that's fun. And in putting all the pieces together to actually make it a reality, that's fun. That's sort of what I wanted to do.

Elad Gil

I think there's been a really interesting trend over the last couple of years, too, where I feel like there's many more founders of this founding generation who want to do the next act. And I think that's a really interesting change from history. And there's other people who've done things like that over, you know, like, Brian Armstrong is working on a biotech as well as running Coinbase. Like, I feel like there's lots of examples now of people saying, I really want to do extra stuff in my life. And that level of ambition, I think, is really impressive.

Because I think it's something that before you'd kind of run a company and if you'd leave, you'd kind of go and retire. There were a lot of almost like, retirement people from the 90s. So I think it's a very exciting trend in terms of people who really know how to do things now, trying to go and do something again.

Apoorva Mehta

I actually think that if you think about it, right, you have to ask yourself, is it inertia that's, like, making you stick around, or is it the first principle way of thinking about it? Is this like, what your values are, what you want to be doing with your life?

And I actually talked to dozens of founders and how they're thinking about their second act, and it was clear that this was on everyone's mind. And for me, I love doing just one thing at a time. I think that when you obsess about a problem, you can really figure out, like, you find solutions in a way that maybe giving it a part-time focus would just not be able to do.

Audience Speaker

I remember reading a tweet. I think that you tweeted on IPO day about the story of how you focused on profitability and achieved that. And the story today around what you did after news of the Whole Foods acquisition, sort of like, mirrored, how you communicated to the team and the framing of the being like, this is the key focus.

I guess I'm curious is prioritization what you would distill as the most important thing in key challenges like this?

Where once you have the right priority and communicate to the team, and then you have the team that has the right strength, that's sort of what allows you to prevail in these moments.

Or how would you describe what allows you to fight these challenges successfully?

Apoorva Mehta

Yeah, I actually think a small focused team versus a large team focused on many different things. I would bet on the small focus team every day. And that's why that's like, one of the biggest advantages of early-stage companies. You have this team that is so focused on this one problem, while a larger company may have like, 15 different product lines.

And so there is this incredible quote which is, never waste a good crisis. And I've always been inspired by that. And so anytime there is a crisis, you have to ask yourself, well, two years later, where would I like to really be and how can I leverage this crisis for the team's advantage? When you have competition, that could be a very good forcing function for your company, because everyone knows who is the enemy. And that clarity is just so incredible that you're getting it for free. So we loved having that focus that brought to bear the whole company's creativity to solve some very complex problems.

Audience Speaker

Looking back at Instacart's journey, what do you view as the major inflection points? Were they expected or not?

And then the second question, what do you regret most?

Apoorva Mehta

We had quite a few inflection points. First was when we signed the deal with Whole Foods that sort of validated that our model was actually something that could work. All right. And most people don't know this, but we had a terrible deal with Whole Foods and there was no math you could ever do that would ever make sense. I talked to the board about it and they were like, are you sure you want to sign this deal? And I was like, yes, we're going to sign this deal and it's a one-year deal, but we're going to sign it. We're going to prove to them that this actually makes sense. And our next deal was three times better.

And that legitimized what Instacart was to other grocers, and that allowed us to get the next deal. And the next deal, that was a massive inflection point for us. It'd be hard to say that COVID was not an inflection point. Of course, overnight we became a lifeline from being a convenience. And that was just incredible. The way that the company handled it. In terms of regrets, I'm sure there's many if I step back and really reflect on it. I think that I'm just really proud of the way the team handled some of these inflection points, and stepped up at the right moments.

Elad Gil

I think Apoorva’s biggest regret is not spending more time with me during that period. Next time, next company.

Apoorva Mehta

There you go.

Audience Speaker

I understand, having explained your framework for accepting bad ideas or proving something that's right to pursue. But before that, going back to you pursuing 20 different things in a couple of years versus more recently, taking a step back more than six months to assess markets as well as landscape, how would you say your ideation process has evolved? And do you have a framework for that as well?

Apoorva Mehta

Yeah, so the first thing I'm looking at is the catalyst. What has changed in the industry that is very meaningful and this is actually very hard to get to? Of course, when mobile came around, you're like, okay, great. You would have to be pretty silly to not see mobile right, like, when it was around as a disruption. But actually to understand the catalysts in a different industry, you have to go deep in the industry to really know what are the regulatory changes, maybe there are some other structural changes that people are not talking about.

For example, one thing that people forget between Webvan and Instacart was credit card penetration had increased very dramatically. So now e-commerce was a thing. Commerce couldn't have been something that was successful in the Webvan era. Right? And so there are all these catalysts that are not something that you see in a headline. So that's number one. First thing you're looking at in any industry is catalysts. What are the catalysts that are changing? And then after that, when you're evaluating a business as a founder, maybe this is different for an entrepreneur or as an investor.

The things that I'm looking at are sort of threefold. Number one, is it a good business? Obvious to say as an investor, but you'd be surprised. Number two, is it going to be something I'm going to enjoy doing? I'm going to have fun? And number three, is it a mission that I can get behind?

The advice that you normally get is the opposite of that. Oh, is it a good mission? Is it going to be something fun? And is it a good business? Right. But the reality is that if it's not a good business, you're not going to have fun doing it and you're not going to achieve the mission.So you sort of need to sort of reverse it as a founder. Now, for each one of these categories, I looked at climate, space, robotics, a couple of other areas, and I was just trying to look at that. Are there really good businesses that I could have fun doing, that have a mission that I could rally behind?

MY BOOK
You can order the High Growth Handbook here. Or read it online for free.

OTHER POSTS

Firesides & Podcasts

Markets:

Startup life

A brief guide to startup pivots

4 Ways Startups Fail

Founder Investors and Scout Programs

Co-Founders

Raising Money

Fireside chat with Satya Nadella, CEO of Microsoft

Elad Gil — Wed, 08 Nov 2023 13:43:05 GMT

Stripe hosted the below fireside chat between myself and Satya Nadella, Chairman and CEO of Microsoft.

We discussed:

Evolution of Microsoft from 1990s to today
AI & MSFT
Societal impact of AI (healthcare, education etc)
AI Safety
Growing Microsoft from ~$250B to $2.5 Trillion & how Satya thinks about his legacy

Video:

Transcript:

Elad Gil

Thanks, everybody for joining today, and thanks again to Stripe for hosting. It's a huge pleasure and honor to have Satya Nadella with us here today. He is the goat of technology CEOs. So, it's like having LeBron here with me or something similar is sort of what it feels like.

Just for a little bit of history, you joined Microsoft, I believe, in 1992, ran a variety of different divisions, and took over as CEO in 2014. When you joined Microsoft, it was perceived as sort of the behemoth of tech. And then it went through a period in the 2000s where it was viewed as a company that was still very strong in enterprise and other areas but wasn't necessarily going to set the stage for the future.

And then I feel like under your tenure, it's transformed into one of the most exciting and interesting companies in all of technology. And that's through big moves around LinkedIn, OpenAI, gaming, just a variety of different areas.

It'd be great to just hear first, what was Microsoft like when you took over?

And how did you think about the culture, the directions, the strategy, and how you were going to impact that over time?

Satya Nadella

First of all, it's great to be with you, Elad. And it is true. I've now spent over three decades in one company, and it's pretty much all my professional career, and I have better language to describe it now. But I think companies, I mean, here there are lots of folks who are founders of companies, and I think companies also go through these refounding moments. I picked this up from Reid Hoffman, which I think is a fantastic frame. It's not talked about as much, although in 2014, when I became CEO, having grown up in the company as a consummate insider, I felt that we needed to essentially refound the company, ground ourselves back in what's the core sense of purpose and mission so that then we can pursue something new and bold.

Keeping that mission in mind. In fact, when I joined in 92, we used to talk about a PC in every home and every desk as our mission. In retrospect, it was easy. You could put it on a spreadsheet and calculate it and what have you, except that by the late 90s, we have more or less achieved that, at least in the developed world. And I felt ever since then a little like, okay, what's our mission? Is it mission accomplished? Is it time to return all the money to the shareholders?

And so that's why I went back all the way, quite frankly, to the very genesis of the company. After all, we were a tools company first. Bill built the basic interpreter for the Altair. And I said, god, there's more need for tools and platforms in 2014, 2023. I'm coming from OpenAI's Dev Day. That's who we are, right at the core, we build tech so that others can build more tech. Let's ground ourselves in that mission. And that has been very, very helpful.

The other one, though, is culture, is the other piece you brought up. I distinctly remember at Microsoft in the late 90s, walking around campus and there were just all the folks sort of saying, God, we must be God's gifts to mankind because we are so good now that the market also recognizes it, except that it was obviously the beginning of the end in some sense, right? The day hubris takes over and you're not grounded in what made you, first of all, successful.

That customer feedback loop, that hunger to learn, to experiment, to do things, which startups do. And so I picked up this cultural frame from Carol Dweck at Stanford around growth mindset, and I said, God, we don't want to be these know-it-alls, but we want to be learn-it-alls. And that has been a godsend, right? Because, one, it is not considered new dogma from a new CEO. It is something that I think spoke to people as humans. It helps us be better parents, better partners, better colleagues, neighbors, and leaders.

And so between these two things, that renewed sense of purpose, I tell you especially, you shouldn't. I mean, founders have this innate capability because they've created something from nothing. And then there's the cult of personality that carries forward. But at some point, if your company is going to outlast you, that mission has to be a lot more than a cult of personality. And so that's why I think CEOs in particular, or mayor model CEOs like me, being much more focused on what is the mission, what is the culture, is the one thing that I would say are two critical things. And of course, that gives you the permission to then make the right calls, but you have to make the right calls on strategy, picking, and execution, but at least it gives you better shots on goal.

Elad Gil

How do you think about the ability to pursue strategy in today's climate?

So I know M&A is a little bit more challenging for a variety of reasons. There's a big wave of AI and other technologies sort of happening.

What advice would you give to founders in the audience or people who are running successful businesses in terms of how to think about their directionality?

But also, how to think about tools like M&A or organic growth?

Satya Nadella

We talk a lot about M&A but I think all of us fundamentally bet on organic growth, whether you're a startup or you are a large company. If I look at, even at Microsoft, I think a lot of what gets written about is our inorganic.

But when I think about most of the big hits and most of the big revenue generators are quite frankly partnerships. And organic partnerships are another thing that people don't talk about. The OpenAI thing is a great partnership. I grew up, in fact, the Gates Grove model is something that I love, which is Intel and Microsoft created the PC ecosystem. It was one of the most wonderful partnerships.

It's also important in partnerships to know what happens if one of the partners becomes too greedy, then it's unstable. But if you can cultivate great partnerships, that's another fantastic source of growth for companies. And so yes, inorganic partnerships and then strategic M&A all three matter for a company today.

I think we're going through a little bit of what I'd say is regulatory adjustment around the world is the best way I can describe it to understand whether we should allow M&A to big companies, just because they're big, shouldn't be acquiring, which I think will have a chilling effect. And it's not exactly good for quite frankly, more startup creation and more vibrancy. But let's say that settles. But I think yes, I would always look at all three of them as organic.

First focus on having a great organic plan because that's the one thing you can always control. Partnerships are something that I would say, don't think of it as a press release, but I think it can absolutely be. And long-term stable partnerships, where you win, I mean it's the three wins, right? The customer wins, the partner wins and you win. In fact, the best partnerships are when you start thinking, caring about the partner, really making sure that they're getting economic surplus.

Sometimes in our Silicon Valley culture, we have a little bit of excessive zero-sum stuff. There are very few zero-sum battles, actually, when you think about it. But we are very obsessed about everything as being zero-sum. And that's where I think a little bit of subtlety, and nuance would help.

Elad Gil

So one of the big waves that's happening right now that Microsoft has really been central and seminal to is this change in AI. And that's transformer-based models and diffusion-based models have changed the trajectory of what we can do with these sorts of machine systems.

You've lived through some of the biggest technology waves of the last 30 years in a really central position. And that's everything from the Internet to the Cloud to the PC revolution, which you mentioned, and mobile.

How big of a deal is AI in your mind relative to all these other trends?

Satya Nadella

I think it's sort of as big as any one of the things that you mentioned. The way I try to come back to what exactly changed and how we should relate to it in terms of any business building, product building. I think there are two big changes.

One is I think we're going to think about application interfaces drastically differently. I mean, we've been talking about natural user interfaces forever, you could say, for 70 years of computing history, from Engelbart down, has always been about like, hey, how do we have the ultimate human-computer symbiosis? And I think we now have some new tools to rethink that. It starts with chat, it starts with text, but it goes quickly beyond that. It's multimodal and it's going to be very interesting for us to sort of apply.

So that's why in some sense it reminds me a little bit, at least in our history, of what happened. And I mentioned this to you backstage when I joined Microsoft in 92 just released Word or Excel or PowerPoint or Publisher. Each day there was a new app. And because Windows 3 was just happening, it feels like that, which is how I can rethink even existing categories with a new interface. But we also know that new platforms are not about just taking the old and just building a new UI.

But there is also, what is it this UI can create as a business that didn't exist. That's the thing that I'm most interested in. Mobile had a lot of things we did on the desktop, but it also created companies like Airbnb and Uber. And I think that that's something that we should think about on the UI side. What's the UI for the AI first app? I don't think we've yet cracked it, but I think we're getting close.

The second thing that I also think is more, in fact, the other technology that people don't talk about as much. I don't know why. When we talk about our big paradigms are relational databases. God, what a thing it was. It is like the other secular journey in digital technology was digitizing people, places, and things, right? That's another 70 years. All we do each day is we wake up and there are more places, people, and things that are digitized, and relational algebra and relational databases helped us reason over it in interesting ways.

I feel like we now have a new pattern recognizer or new reasoning engine in this, doing neural algebra on it. So I feel these two things, Elad, if you take an existing category, I have a new way to think about the user interface. I have a new way to reason about all the data I have and the knowledge in the world and do the join, so to speak. If you start thinking of applications that way and then building a new to the world business, then I think this would be as big as at least any one of the things that you mentioned.

Elad Gil

A lot of the trends that people are talking about right now that at least strike me as a little bit early, but may be very interesting in the future, that captures both of the things that you mentioned is moving to an agent world or an agent-driven world. People talk about how eventually we'll have agents that will represent us, will represent apps we interact with, companies, governments, etc.

How much of a believer are you in this sort of future world of agent-driven action or agent-driven interfaces?

And what do you think the timeline is for that if it does exist?

Satya Nadella

I'm a big believer in that. In fact, at least in our world, the design pattern I love a lot, which we picked for our own product building, which we evangelize as a pattern for anybody, is this copilot.

First of all, there's a human in the loop. The human has agency and judgment of human matters. And so the first instances were things like GitHub Copilot, where the copilot is built into the app canvas, so to speak. There's a sidecar, and those things all work together to help you with your task at hand, which is to get your coding done.

And then we have now propagated that into knowledge work with the Microsoft365 copilot across all of our surfaces. But ultimately, when I think about what we did with Bing Chat, I think of Bing Chat as basically the web copilot. This M365 is the work copilot. So the web Copilot and the work Copilot could be sort of the universal agent, so to speak. That's kind of how at least I conceive of it.

But it needs one important capability, which is it needs to be able to talk to other agents, a customer service agent, a travel agent, to get work done. Some of it’d be interrupt-driven, in the sense that it'll be autonomous but it'll also bring back to you for decision-making. So I think that one of the key runtimes of our time will be that multi-agent runtime.

And we have a thing in open source called Autogen that is getting some good traction. So we are building some of the stuff similar to that underneath what is our copilot. You know, OpenAI launched a bunch of very interesting stuff with GPTs, which is sort of, I would say, early agents on top of chat, GPT itself. They even have an Agent API, all of that we will put into our copilot ecosystem.

So I think yes. So to your fundamental point, I think this idea that people will have agents, these agents will interoperate with each other, there will be some type of super apps that whoever cracks, there'll be a few runtimes where naturally people will gravitate to, which will be these multi-agent frameworks.

Elad Gil

You mentioned as well that you're very excited about the business potential of a lot of these things and what's going to transform. And obviously, Microsoft has done some very innovative things through GitHub Copilot, through some of the other areas that you mentioned.

Are there any areas of the business world that Microsoft is not directly involved with that you think are most likely to transform via AI in the near term?

Satya Nadella

That's interesting. If I knew about it, I probably would be in it, but one of the things I am grounded on is I think the time has come for us as a tech industry to directly parlay what we celebrate as tech advances into broad economic growth.

Because I think there's this real critique, which some of it is, I think, well-founded about. Hey, you guys talk a lot about sort of all this tech, but where is the economic growth? I mean, last time we checked in the developed world, inflation-adjusted, we are probably at what, zero or negative growth. And so there is a real need for economic growth and economic growth that comes while keeping the planet healthy and more equitable in terms of the growth itself. So there are many other things as well, which I think are core responsibilities. But that said, we have to drive economic growth.

And that's where I got excited with GitHub Copilot, right? When we can take something like software development and one, bring joy back to software development. Man, what we had done to it, the fact that I've just got to copy-paste from a variety of places, and get distracted. Instead of that, let's just stay in flow focus and then see how you can bring back productivity to the software developers.

But the interesting thing about the more we study GitHub Copilot's effect on an organization is if you get software development to be faster, all the other functions around change. The workflow, when a salesperson does a pull request, that's kind of my thing about like, wow, that's a different org. And so now you can imagine an entire organization that's moving at a different pace.

So I think we are at the frontier of figuring out what it means for both having these productivity boosters inside of every function and then what is the workflow if you have it like you talked about, agents, I think we will have to discover.

If you remember in the mid-90s when we first started putting things like systems of record like CRM and ERP, we used to talk about something called business process reengineering. It was as much about a new methodology or a doctrine even of how you run businesses. Like you didn't have five finance departments or you didn't have manufacturing cogs, not in financial accounting. So some of the business practices will have to change.

So that's why I think what is probably interesting would be how people think about a vertical industry or a business process probably is going to be very different.

One of the things I think most people think about is “Can there be a large business created that only focuses on one vertical or one business process?”

I think there can be. It all depends on how much economic surplus you can create using this technology.

Elad Gil

You mentioned partnerships earlier and how they're a key linchpin to Microsoft's ongoing strategy and how it's worked with different partners over time. One thing that I think has been very exciting is how you've both worked with OpenAI very deeply for their closed-source models as well as Meta around Llama and the open-source world.

How do you think about how Microsoft engages with open-source and closed-source AI evolves over time and those relationships with those partners in particular?

Satya Nadella

Yeah, I mean for us it's not an either or. GitHub exists primarily because of the permission to support the open source ecosystem, so it's sort of not something that we take lightly, and therefore we will always be as a company. In fact, it turns out I think most people don't recognize we are one of the biggest contributors to open source. We are probably today the biggest contributor to Linux as Microsoft. So it's sort of now fundamentally ingrained.

But that doesn't mean we don't have a bunch of proprietary closed-source systems and revenue streams as well. So, to us, it's pretty much part of the business. Then the question is, what's the best way to be able to meet the developers, meet the companies, and meet the organizational needs that we serve well?

And so that's why even when it comes to foundational models, obviously OpenAI is our lead partner when it comes to frontier models, but we have a lot of open source models, including ours. We're really excited about even our own contributions. Like I love these SLMs, the small language models with PSI. And so we're going to always contribute. We'll support and make sure that developers have a choice. And in our own products, you'll see a mix of use as well.

Elad Gil

One of the emerging critiques around both open source models as well as advanced LLMs is concerns around safety. And I think people often mix three types of safety.

There's safety that I'd almost view as textual safety. It's misinformation, it's bias, it's hate speech, etc. There's physical safety, will the AI be used to derail a train or develop a virus or things like that? And then there's existential safety.

Do you worry about at some point having some sort of confrontation with an AI species or intelligence that's gathering resources or whatever it may be?

How do you think about those critiques? And how do you think about AI safety over time?

Satya Nadella

Yeah, I remember talking to you a few months ago and you had laid out this taxonomy, which I like. I think one of the things that we don't have is a well-grounded way to talk about these three levels. Because when we say safety, it could mean, wow, we are talking about an existential issue or election interference, deep fakes or what have you. So I think let's unpack them that way.

So my feeling is that the first thing we've got to really go and focus on is any real-world harm today, because of any AI deployment that sort of has not gone right. In fact, in democracies in particular, the thing that I worry the most about is, quite frankly, elections and our democratic process somehow being unduly influenced by some use of AI. I think that's the place where rightfully so, we'll be held to account if something goes wrong there. Because after all, that was the critique of what happened in social media. Everybody was excited about social media because of the Arab Spring, but it kind of nearly broke democracy. And so therefore, now everybody says, look, we're not going to let that happen again.

So the question is, what do we do there? What does the government do? What does the civic society do? What do companies do? And quite frankly, I think we somehow say that this is, at the end of the day, it's a societal choice. It's not any one company can do everything here. It's sort of a societal choice.

Like, for example, in the United States. After all, you have to be able to come together and say, how do we think about free speech? And what turns out to be free speech that is now bordering on election interference or what have you, or disinformation? These are tough things, and so therefore, this is not like a decision an AI can make or a decision a company's moderators can make. They're societal norms and decisions. And so that's the most complicated process we have to engage in, quite frankly.

The other ones, the existential risk one, there's time. Second, I think there are more engineering solutions to that in an interesting way. The fundamental thing is if we have a self-improving program that we've lost control over, the last time I checked, there are people in other engineering fields who know control theory and how to think about it. So there must be stuff we can learn from others and apply it. But I'm glad that the dialogue is happening on the existential risk at the same time.

The other one in the middle, which you talked about, which I like, which is before we deploy AI into the real world. Like, for example, if it's going to actuate stuff in the world, that might be a thing where you may want to think about more carefully.

And that's where risk-based regulation is also a helpful thing. It's like cars do get deployed, but they have regulations that are different than a lot of other regulations. And so we have in healthcare, in financial services, in auto safety, already existing regulations, and we can subject AI to the same regulations. So I think having maybe for the middle, more risk-based approach, more research funding, quite frankly, for the existential stuff. And then for the real-world harms today, whether it's bias or election interference, both: What are the societal norms that are as much about us as a democratic society, and especially in the United States? I hope we take that responsibility, and I really hope there will be an election in which a politician wins the election where they're able to articulate a vision. This is where it bugs me when tech and tech leaders are the ones who are talking about this. Right? I didn't elect a tech leader to speak for me as an elected official. So, I would rather have the representatives of the people win elections by having a real vision for what the norms are for how we engage.

Elad Gil

Related to that, there's a lot of promise for AI in terms of global equity. And so when I look at the really big thrusts in terms of where it can impact things, one is in healthcare, and you see things like the Med-PaLM 2 model from Google, which has enormous performance on the healthcare side and the ability to have good output relative to health-related criteria.

And then education. And Microsoft owns Minecraft, which is one of the most popular tools that I've seen parents engage with. And when I think about the ability to impact education, it's going back to this agent-driven world of something that can customize or interact with your child in a rich way, can tutor them, can teach them.

I'm curious about your thoughts on either those areas of global impact of AI in a positive sense or other areas that you're most excited about.

Satya Nadella

Yeah, I think education, health, and perhaps financial inclusion, would be the three things.

The first one on the education side. I was in the UAE just last week and it was fantastic to see they have an AI minister in the UAE in their health department. They launched for the country an AI tutor built on one of the GPT family APIs. But also I met the founder who was building that and using a variety of different techniques to sort of ground that model. The beauty of having a personalized tutor for all the students of the world is absolutely imminently doable. When you think about the government transfers, and government subsidies that exist in education, all over the world, even in the developing world. And now to be able to have GPT4 turbo pricing, I think it's easy for us to think about delivering personalized tutors that are very, very capable. So I'm very excited about that.

Healthcare is another one. Like if you take even the US, what, 19% of our sort of GDP is in healthcare, we know we need the health outcomes to get better and costs to come down. And a lot of the cost is quite frankly, workflow cost, right? It's not even like we always go to some magical drug, this, that, and the other. Those are all fantastic. It's important. But the real cost is in care management. And that's why I'm excited about partnering with someone like Epic who can make a massive difference. Epic is everywhere. They're ubiquitous, they're building generative AI deeply. And I mean these things like they were telling me about all these use cases when you finish a shift and you want to hand off that process. Why don’t I have a summary? Why don’t I, when I am in the bedside, have a bot that I can interrogate to ask all the questions and not have anything lost in the inbox for the physician? With Nuance, we have this fantastic sort of way now to be able to really transcribe their doctor-patient conversation and reduce physician burden. And that technology, once built, the software once built can reach every hospital in the world.

And so those would be the two places I think that we can make a huge difference.

Elad Gil

We talked a lot about AI. Microsoft obviously does an enormous amount of things in an enormous amount of important areas relative to B2B, consumer, etc.

What other areas are you excited about besides AI?

Satya Nadella

We just finished closing Activision Blizzard. We are very excited about gaming. You brought up Minecraft. So, gaming is another place where I think is a category for us. When I look back at Microsoft, there are three things that I think come naturally to us, which is in our, I would say DNA.

One is these platforms and tools. We will always be a platform developer tools company. That's sort of the core heritage. The second thing is productivity and communication software. I think that's the other one that we do. And the third is gaming. In fact, I think Flight Simulator was built before Windows was built. And so we will always be in all these three categories.

And also in gaming. What is exciting to me is we love the console, and we love PC gaming, but with some of the changes in cloud gaming, we can get gaming everywhere. And so that I think is going to be the place where AI as a platform for all of these would make a real difference as well.

Elad Gil

So I think you've had one of the most successful CEO runs of all time in the history of capitalism. I don't say that lightly. And you took a company that was worth 250, 300 billion dollar market cap and at the time people thought that was an insanely high market cap. And now you're at 2.5 trillion. So you've added 2.25 trillion in market cap over the last eight or nine years.

What do you want your legacy at Microsoft to be? And how do you think about that going forward?

Or how do you think about it years from now and you're looking back, what do you want to have accomplished?

Satya Nadella

Yeah, I mean, the way I think about legacy perhaps, I'm very suspect of anybody who comes in and says, oh, I showed up at the job. The person who was there before me was all terrible and I changed it all and it's all me. I'm very suspicious of those people because at some level the job is to build institutional strength so that the people who take over from you are more successful than you.

In fact, I always think about the next person, and by the way, I always thought about this growing up in the company, that if the person who takes over from me in any role is able to succeed, then maybe you did something right. So quite honestly, in tech, there's no franchise value. We know that you have to sort of reinvent yourself. So that means some of the things you do, you have to also really, like somebody said, the metaphor of you got to make sure you leave enough energy in the system so that it can continue to renew it.

Just thinking of success by market cap sometimes is definitely not the way to measure things because you got to invest like long before it's conventional wisdom. You can't expect markets to always reward you. In the long run, they will. But, there will be periods of time, as Bezos would say, you got to be okay being misunderstood, which I think is right. So therefore I look at all of those as perhaps long-run indicators or something that went right, but not the only indicators.

And so therefore I'm much more focused on making sure that Microsoft is doing relevant things in the future. And if we have created enough institutional strength, cultural strength, we're not doing things out of envy. We're doing things that are useful in the world. We're doing things that sort of help the company succeed. I feel if the world around us is doing well and we're doing well, that's a fantastic sort of equilibrium to achieve.

Elad Gil

Makes a lot of sense. And then the very last question because we're almost out of time. Thank you again for being so generous with your time.

When you look at other areas of technology that you consider very important, there are a lot of different things people are talking about now in terms of fusion or other forms of energy, self-driving cars, and how that's going to transform cities and transportation. There's a variety of these areas.

Are there any that you're watching most keenly or that you're most excited about the societal impact that they may bring?

Satya Nadella

I think energy. I mean if I look at it, even for us, it's pretty existential, right? If I look at our own energy needs, as we think about AI and our build-out, we definitely need it. I mean, at some level, the lucky break we have is it's all fresh-project starts. We're the largest purchaser of green energy today in the world. So that means we can stimulate even a lot of these new projects, which could be sometimes risky.

But I feel the biggest thing that we have to make is the energy transition. And the energy transition is a complex thing. I mean, I had never realized this effectively. I think what we are talking about is you have to take 250 years of chemistry and the entire petrochemical sort of value chain and sort of somehow compress it into 25 years. And I look at that and say, wow, that's the real challenge.

Even from an AI perspective. I'm excited about this AI stuff being sort of super useful in somehow synthesizing some new molecule that then helps with some new batteries, which then help with taking the abundant solar power and making it much more possible for us to tap into it even in a data center.

So those are the kinds of real problems that need to be solved quickly. So that's the one industry more so than anything and that entire chain that I think we have both an adjacency to it and a real dependency on it.

Elad Gil

Well, fantastic. Thank you again for joining today.

Satya Nadella

Thank you so much Elad. It's been a pleasure.

MY BOOK
You can order the High Growth Handbook here. Or read it online for free.

OTHER POSTS

Firesides & Podcasts

Markets:

Startup life

A brief guide to startup pivots

4 Ways Startups Fail

Founder Investors and Scout Programs

Co-Founders

Raising Money

Unicorn Market Cap 2023: Rise of AI

Elad Gil — Wed, 20 Sep 2023 18:01:56 GMT

[This post co-authored with Shreyan Jain]

We’ve previously explored the geographic distribution of global unicorn market cap in 2019, 2020, and 2021. In the past, we’ve used this analysis to highlight trends like the concentration of tech startups in specific industry-towns, a slowdown in new unicorn creation in China, and the prevalence of post-Covid remote work policies in the Bay Area.

A lot has happened since early-2021, including public tech market and multiples highs, ZIRP (Zero Interest Rate Policy), inflation, interest rate hikes, and a massive public market stock valuation correction, so now is a good time to re-examine what unicorn market cap looks like.

Big Takeaways / TL;DR:

Despite the “you can build a startup anywhere” mantra during COVID, unicorn market cap continued to aggregate in the USA (53% of global market cap). Indeed, 40% of all the unicorn market cap in the world continues to be found in just 3 US cities: SF Bay Area (26%), New York (8%), and LA (6%).
The wide distribution of cities with just 1 unicorn (up from 37 to 75 in 4 years) likely reflects a rise in “ZIRPacorns”, to coin a term, rather than a true decentralization of important startups. ZIRPacorns are unicorns that would probably be worth dramatically less than $1 billion if not for the COVID/ZIRP-era multiple excess.
SF has emerged as the center of AI unicorn market cap with an 81% initial market share of generative AI companies. It is, of course, quite early days and this will likely evolve with time.
NY has grown global unicorn share from 5% to 8% since 2020. Of the top 10 unicorns in NY, 6 are crypto and 3 are fintech, suggesting that NY is a fintech/crypto cluster to date morphing towards a general purpose ecosystem. LA also grew its share during COVID - largely due to a mix of SpaceX and other defense or space companies like Anduril and Relativity. These 3 companies make up 67% of LA market cap.
Paris and London also gained share, while China has continued to lose share over the past few years.
A few cities or regions are highly dependent on single companies for much of their market cap. For instance, Bytedance makes up 58% of Beijing’s total market cap of $387B, SpaceX is 62% of LA’s $222B, and Shein is 64% of Shenzhen’s $157B.
Miami, Austin, and others have many Barry’s Bootcamps, but fewer unicorns than popular perception might suggest (see chart). We look at the Barry’s/Unicorn ratio of key cities. SF Bay Area is dominant with a unicorn/Barry’s ratio of 48. This is of course a lagging indicator and we think Miami in the long run may benefit from talent turnover coming from NY, much as LA was by the Bay Area. Alternatively, the Bay Area may become more fit and reduce this ratio.

Caveats

All raw data was taken from CB Insights and can be found here. Some caveats:

The data for the 2023 batch of unicorns was downloaded on July 20, 2023 and may not include all recent unicorns.
The listed market caps are based on last-round valuations, many of which are inflated 2021 numbers and may not reflect current fair-market value.
Unicorn market caps are inherently backwards-looking and may not adequately capture trends that we’re still in the early stages of, such as the market reset for growth-stage companies or the proliferation of generative AI.
There are undoubtedly errors in market cap, mismapping of a company to a city, etc across multiple years of data. That said, major trends should hold.

New Unicorns

The number of unicorns has exploded from 44 in 2012 to 1215 today—a nearly 30x increase in roughly 10 years. Going back to 2019, the number of unicorns has roughly doubled every 2 years, with new unicorns in any given period outpacing existing unicorns that “graduate” via an IPO, acquisition, or downround. Since June 2019, 6 times as many private companies become unicorns (1094) as the number of unicorns that became publicly traded companies (172).

In both 2020 and 2021, new unicorns accounted for roughly 40% of all unicorns; that rate has shot up to 52% today. This trend becomes even more apparent when you look at decacorns; companies that crossed the $10B mark within the previous years account for 60% of all decacorns today, up from 35% in 2021.

New unicorn creation in the past 3 years spiked in 2021 and early 2022 before falling back to pre-COVID norms in the last year (it’s likely the data here is off by a few months, since there’s a lag between when fundraising rounds close and when they get publicly announced and recorded in datasets like CBI).

In 2020-2021 interest rates dropped, stimulus was pumped into the economy, and both public and private valuations soared. Assuming private market valuations followed public ones, many private companies were overvalued by 3-4X at their peaks. This suggests a threshold of $3 to $4 billion market cap at time of funding in 2020-2021 equates to a “normal” $1 billion company in non-ZIRP times. Indeed, when we compared 2019 (at $1B) and 2021-2023 (at $3B) vintages in terms of new unicorn distribution, we saw nigh-identical patterns.

Since COVID there has been a large reset in public market valuations with many technology companies losing 50-90% of their market caps. Private markets adjust slowly, with markdowns, firesales, and shut downs often going unreported. We would anticipate many of the unicorns from the last few years will lose their status as $1B+ companies in the upcoming years.

Unicorn Creation by Region

The US drove the majority of new unicorn creation in the last 2 years, accounting for 56% of brand new unicorns by count and 57% of new unicorn market cap. Although China’s velocity of new unicorn creation has bounced back a bit since a decline in 2021, it still lags behind the United States and has now been overtaken by Europe, with India nearly catching up as well.

Some city-specific trends:

SF Bay Area and New York continued to grow or maintain their global unicorn market cap share. SF stayed level at 26% from 2021 to 2023, while NY grew its share from 6% to 8%.
London saw its unicorn count double from 2021 to 2023 and has continued to be the primary tech industry-town in Europe. Paris and Berlin are the next 2 major drivers for new unicorns in Europe.
Beijing still accounts for more unicorn market cap than New York in absolute terms, but New York added 6 times as many new unicorns and 8 times as much new unicorn market cap in the past 2 years.
Tel Aviv has grown rapidly despite the small population of Israel; it added more unicorns in the past 2 years (15) than it started off with in 2021 (10), and has nearly tripled its unicorn market cap.

Unicorn Concentration by Region/Country

When we look at total unicorn numbers rather than just new unicorns created since 2021, the US’s dominant position becomes even clearer. The US has actually increased its global share of unicorns – in terms of both count and total market cap – over the past few years. Every other region also saw its share of total unicorn count and market cap increase except for China, who went from housing 26% of all global unicorns in 2019 to just 15% today, and from 30% of global unicorn market cap down to 20%. Europe has almost closed the gap, and in fact already has almost as many unicorns (165) as China (178).

When we break regions like Europe, APAC, and LATAM into their respective countries, we can see that 82% of all unicorn market cap is concentrated within just 4 countries: the US, China, India, and the UK. Every other country accounts for <= 2% of global market cap.

Unicorn Concentration by City

One of the most interesting consequences of ZIRP has been a proliferation of unicorns outside of major cities. In particular, the number of cities with exactly 1 unicorn has doubled from 37 in 2019 to 75 in 2023, and the number of cities with <= 2 unicorns has also >doubled from 48 to 102:

The current day distribution of cities by their unicorn count clearly shows this fat tail of “one-hit wonders”:

Although this has led to some diffusion of global unicorn market cap, most value continues to be concentrated in the tech clusters on the right side of this graph; the top 15 cities across the world account for almost 75% of all unicorn market cap.

Outliers & Ecosystem Dominance

There are 3 companies – Bytedance, SpaceX, SHEIN – with market caps above $100B; The next largest unicorn is at roughly $50B. These three outliers make up a disproportionate share of not only global unicorn market cap (12%), but also the unicorn market cap of their respective cities (Beijing, Los Angeles, Shenzhen) – around 60% in all three cases.

In a way, these 3 outlier cities take the “power law” of VC to an extreme—they are home to a fat tail of low market cap unicorns, but virtually all the returns are concentrated in the biggest “winner”.

We view dispersion of market cap across many companies as a broader signal of startup ecosystem health. For example, the largest companies in both the SF Bay Area (Stripe) and New York (OpenSea) only make up 5% of their respective regional market cap.

US clusters

The top 10 cities in the US make up almost 90% of its total unicorn market cap, led by the Bay Area, New York, Los Angeles, Boston, and Seattle (same order for top 5 by unicorn count, except for Boston and LA flipped). LA is dominated by aero/astro and defense companies, with SpaceX, Anduril, and Relativity combining to make up two thirds of total LA market cap. Meanwhile, Boston is dominated by biotech companies—home to 30% of global biotech unicorn market cap.

New York is a fintech and crypto cluster

Looking at the top 10 unicorns in New York today, it’s clear that its ecosystem has specialized in fintech (9 of top 10) and crypto (6 of top 7). Of the 10 most valuable private tech companies (on paper) in NY, all 10 became unicorns for the first time between July 2020 and November 2021. A number of these ran up from $1B to $6-13B in market cap during ZIRP and COVID. A subset will be sustainably valuable; for example Ramp recently repriced to $5.8B—still a great outcome when public market companies are down 50-80%.

SF Bay Area is the early winner and cluster for generative AI

Although we’re still in the earliest days of what’s looking like a monumental wave, it’s becoming clear that the Bay Area has emerged as an early leader for generative AI companies, with a 81% market cap share. This should be a strong tailwind for the US broadly and the Bay Area specifically to continue growing their share of global unicorns. It is, of course, the very early days of generative AI so we anticipate more global market cap emerging over time.

Note: Below are the companies we considered to be “generative AI” companies for the sake of this analysis; this includes a few GPU companies (Cerebras, Graphcore, TensTorrent) who have enjoyed record growth due to the recent compute demands imposed by foundation models and other gen AI providers. Excluding them would only strengthen the Bay Area’s total concentration.

Unicorn Market Cap Vs Barry’s Bootcamp Locations

A popular narrative over the last few years has been the emergence of Austin and Miami as tech centers. While both regions hold promise for future growth, both are still smaller contributors to new unicorn formation as of today.

Austin has the 7th most unicorns in the US and the 10th most market cap, after Chicago, Jacksonville, and San Diego (the latter two may reflect biotech market cap, versus tech market cap). Miami is home to 7 unicorns worth $15.8 billion, just behind 10th place San Diego (8 unicorns, $30B in market cap).

Comparing unicorn market cap to the number of Barry’s Bootcamp locations by city shows that the Bay Area has 48 unicorns per Barry’s, while LA has just 3 and Miami has just 2. This is due to a higher number of Barry’s per capita in Miami and a lower number of unicorns. NY and Austin both net out at about 15-17 unicorns per Barry’s. Overall this suggests that perhaps the Bay Area is out of shape. We will continue to track this important metric ongoing.

Chinese unicorn market cap often dominated by single companies

Beijing and Shanghai account for 3 out of every 5 Chinese unicorns and two thirds of total market cap. Shenzhen’s $157B unicorn market cap is almost entirely driven by a single unicorn, SHEIN ($100B).

India Clusters

India unicorn market cap is centralized by 3 cities – Bengaluru, New Delhi, and Mumbai – which account for nearly 90% of unicorns and nearly 95% of all unicorn market cap.

When we look at these figures over time, we can see that Bangalore and Mumbai continue to grow while new unicorn creation in New Delhi has slowed.

Europe Clusters

Although it has its own centers of unicorn concentration, Europe is unlike most other global regions in that an outsized share of unicorn market cap has existed ongoing outside of major industry-towns. While the share of total unicorn market cap within the top 3 cities is ~75% for the US, ~85% for China, and ~95% for India, the top 3 cities in Europe – London, Paris, and Berlin – only account for ~55% of its total unicorn market cap, a significantly lower share.

This is almost entirely driven by a fat tail of cities with 1 or 2 unicorns; although both the US and Europe are home to exactly 47 cities with at least 1 unicorn, there are 37 cities with 1-2 unicorns in Europe (nearly 80%), compared to just 26 such cities in the US (55%)

Over the past 4 years, London and Berlin have both held steady at around 35% and 10% of total European unicorn market cap, respectively. Meanwhile, Paris has been on a breakout trajectory, steadily increasing its share of European unicorn market cap from ~5% to ~12% over the same time period, and crossing Berlin to become one of the region’s most important industry-towns in 2021 (more than 80% of Stockholm’s market cap in 2021 was made up by a single unicorn, Klarna).

The End of ZIRP: What happens to the ZIRPacorns?

One hypothesis is that, on average, unicorns outside of industry towns may be less likely to sustain their market caps than unicorns in industry towns. Part of this is the anecdotal belief that VC investors became anxious during ZIRP to find “hidden gems” and therefore funded many marginal companies outside of major tech hubs.

The first signs of the recent turnaround in macro conditions comes in the form of a global slowdown in new unicorn creation. Given the money raised in 2021 by startups, we anticipate it will still be anywhere from 1-3 years before many companies run out of money and need to fundraise/reprice. There are likely 3 long-term outcomes for these companies:

Shutdowns, small exits, and slow burn down. Sycamore Networks, one of the many high fliers of the dotcom bubble, continued operations for nearly 12 years as a public company running off the cash it raised in its IPO. It spent the last 5 years trading roughly at the same value as its cash holdings.

Terminal value companies. Many late-stage companies may be at the highest valuation they will reach in the near run. Many unicorns of today may end up being worth only 20-50% of their current high water mark. They will survive ongoing, but perhaps have reached their terminal valuation early.

Companies that grow out of it. It took nearly 10 years for Amazon to regain its dotcom era high-water mark of around ~$50B after the crash; today it stands at $1.4T, a near-30x.

There is reason for optimism in tech

Despite the gloomy prognosis for many mid- to late-stage private tech companies, there is a lot of room for optimism as well. This includes the aggregate revenue being generated by SaaS companies continuing to scale rapidly (in other words, real value is being delivered by tech companies at scale) and the generative AI wave is remaking how we think about software. There is a lot of room for innovation and exciting times ahead.

MY BOOK
You can order the High Growth Handbook here. Or read it online for free.

OTHER POSTS

Firesides & Podcasts

Markets:

Startup life

A brief guide to startup pivots

4 Ways Startups Fail

Founder Investors and Scout Programs

Co-Founders

Raising Money

Old Crypto Stuff:

AI Regulation

Elad Gil — Mon, 18 Sep 2023 01:34:48 GMT

[While I was finalizing this post, Bill Gurley gave this great talk on incumbent capture and regulation].

ChatGPT has only been live for ~9 months and GPT-4 for 6 or so months. Yet there have already been strong calls to regulate AI due to misinformation, bias, existential risk, threat of biological or chemical attack, potential AI-fueled cyberattacks etc without any tangible example of any of these things actually having or happened with any real frequency compared to existing versions without AI. Many, like chemical attacks are truly theoretical without an ordered logic chain of how they would happen, and any explanation as to why existing safegaurds or laws are insufficient.

Sometimes, regulation of an industry can be positive for consumers or businesses. For example, FDA regulation of food can protect people from disease outbreaks, chemical manipulation of food, or other issues.

In most cases, regulation can be very negative for an industry and its evolution. It may force an industry to be government-centric versus user-centric, prevent competition and lock in incumbents, move production or economic benefits overseas, or distort the economics and capabilities of an entire industry.

Given the many positive potentials of AI, and the many negatives of regulation, calls for AI regulation are likely premature, but also in some cases clearly self serving for the parties asking for it (it is not surprising the main incumbents say regulation is good for AI, as it will lock in their incumbency). Some notable counterexamples also exist where we should likely regulate things related to AI, but these are few and far between (e.g. export of advanced chip technology to foreign adversaries is a notable one).

In general, we should not push to regulate most aspects of AI now and let the technology advance and mature further for positive uses before revisiting this area.

First, what is at stake? Global health & educational equity + other areas

Too little of the dialogue today focuses on the positive potential of AI (I cover the risks of AI in another post.) AI is an incredibly powerful tool to impact global equity for some of the biggest issues facing humanity. On the healthcare front, models such as Med-PaLM2 from Google now outperform medical experts to the point where training the model using physician experts may make the model worse.

Imagine having a medical expert available via any phone or device anywhere in the world - to which you can upload images, symptoms, and follow up and get ongoing diagnosis and care. This technology is available today and just need to be properly bundled and delivered in a bundled and thoughtful way.

Similarly, AI can provide significant educational resources globally today. Even something as simple as auto-translating and dubbing all the educational text, video or voice content in the world is a straightforward task given todays language and voice models. Adding a chat like interface that can personalize and pace the learning of the student on the other end is coming shortly based on existing technologies. Significantly increasing global equity of education is a goal we can achieve if we allow ourselves to do so.

Additionally, AI can also play a role in other areas including economic productivity, national defense (covered well here), and many other areas.

AI is the likely the single strongest motive force towards global equity in health and education in decades. Regulation is likely to slow down and confound progress towards these, and other goals and use cases.

Regulation tends to prevent competition - it favors incumbents and kills startups

In most industries, regulation prevents competition. This famous chart of prices over time reflects how highly regulated industries (healthcare, education, energy) have their costs driven up over time, while less regulated industries (clothing, software, toys) drop costs dramatically over time. (Please note I do not believe these are inflation adjusted - so 60-70% may be “break even” pricing inflation adjusted).

Regulation favors incumbents in two ways. First, it increase the cost of entering a market, in some cases dramatically. The high cost of clinical trials and the extra hurdles put in place to launch a drug are good examples of this. A must-watch video is this one with Paul Janssen, one of the giants of pharma, in which he states that the vast majority of drug development budgets are wasted on tests imposed by regulators which “has little to do with actual research or actual development”. This is a partial explanation for why (outside of Moderna, an accident of COVID), no $40B+ market cap new biopharma company has been launched in almost 40 years (despite healthcare being 20% of US GDP).

Secondly, regulation favors incumbents via something known as “regulatory capture”. In regulatory capture, the regulators become beholden to a specific industry lobby or group - for example by receiving jobs in the industry after working as a regulator, or via specific forms of lobbying. There becomes a strong incentive to “play nice” with the incumbents by regulators and to bias regulations their way, in order to get favors later in life.

Regulation often blocks industry progress: Nuclear as an example.

Many of the calls to regulate AI suggest some analog to nuclear. For example, a registry of anyone building models and then a new body to oversee them. Nuclear is a good example of how in some cases regulators will block the entire industry they are supposed to watch over. For example, the Nuclear Regulatory Commission (NRC), established in 1975, has not approved a new nuclear reactor design for decades (indeed, not since the 1970s). This has prevented use of nuclear in the USA, despite actual data showing high safety profiles. France meanwhile has continued to have 70% of its power generated via nuclear, Japan is heading back to 30% with plans to grow to 50%, and the US has been declining down to 18%.

This is despite nuclear being both extremely safe (if one looks at data) and clean from a carbon perspective.

Indeed, most deaths from nuclear in the modern era have been from medical accidents or Russian sub accidents. Something the actual regulator of nuclear power seem oddly unaware of in the USA.

Nuclear (and therefore Western energy policy) is ultimately a victim of bad PR, a strong eco-big oil lobby against it, and of regulatory constraints.

Regulation can drive an industry overseas

I am a short term AI optimist, and a long term AI doomer. In other words, I think the short term benefits of AI are immense, and most arguments made on tech-level risks of AI are overstated. For anyone who has read history, humans are perfectly capable of creating their own disasters. However, I do think in the long run (ie decades) AI is an existential risk for people. That said, at this point regulating AI will only send it overseas and federate and fragment the cutting edge of it to outside US jurisdiction. Just as crypto is increasingly offshoring, and even regulatory-compliant companies like Coinbase are considering leaving the US due to government crackdowns on crypto, regulating AI now in the USA will just send it overseas.

The genie is out of the bottle and this technology is clearly incredibly powerful and important. Over-regulation in the USA has the potential to drive it elsewhere. This would be bad for not only US interests, but also potentially the moral and ethical frameworks in terms of how the most cutting edge versions of AI may get adopted. The European Union may show us an early form of this.

Regulation can distort economics, drive cost way up, and slow down progress for important societal areas: Healthcare as an example.

Regulation tends to distort the economics of an industry. Healthcare is a strong example where the person who benefits (the patient) is different from the person who decides what to get (the doctor), what can be paid for (the insurer) and what the market can even have to begin with (the regulator). This has caused a lack of competition for many parts of the healthcare industry, an inability to launch valuable products quickly, and in some cases, prevention of adoption by valuable products (due to lack of a payor).

It is telling that during COVID, when regulations were decreased, we had a flurry of vaccines developed in less than a year and multiple drugs tested and launched in anywhere from a few months to two years. All this was done with minimal patient side effects or bad outcomes. This similarly happened during WW2 when Churchill wanted to find a treatment for soldiers in the field with gonnorhea and made it a national mandate to find a cure. Penicillin was rediscovered and launched to market in less than 9 months.

Who do you want to make decisions on the future? Tech people are naive about “regulation”

Most people in tech have never had to deal with regulators. If you worked at Facebook in the early days or on a SaaS startup, the risk of regulation has likely not come up. Those who have dealt with regulators (such as later Meta employees or people who have run healthcare companies) realize that there tend to be many drawbacks to working in a regulated industry. Obviously, there are some positives from regulators when said regulators are functioning well and focused on their mission - e.g. data-driven prevention of consumer harm, in a way that does not overstep the legal frameworks of the country.

In many discussions I have had with tech people who call for regulation of AI, a few things have come out:

Many people working in AI think deeply and genuinely about what they are working on, and want it to be very positive for the world. Indeed, the AI industry and its early emphasis on safety and alignment strikes me as the most forward looking group I have seen in tech on the implications of their own technology. However….
…Most people calling for regulation have never worked in a regulated industry nor dealt with regulators. Many of their viewpoints on what “regulation” means is quite naive.
1. For example, a few people have told me they think regulation means “the government will ask a group of the smartest people working on AI to come together to set policy”. This seems to misunderstand a few basics of regulation - for example the regulator may actually not understand much about the topic they are regulating, or be driven politically versus factually. Indeed, recent example of “AI experts” consulting on regulation tend to have standard political agendas, versus being giants in the AI technology world.
2. Most people do not understand that most “regulators” have varying internal viewpoints and the group you interact with within a regulator may lead to a completely different outcome. For example, depending on which specific subgroup you engage with at the FDA, SEC, FTC, or other group, you may end up with a very different result for your company. Regulators are often staffed by people with their own motivations, political viewpoints, and career aspirations. This impacts how they work with companies and the industries they regulate. Many regulators are hired later in their careers in to the larger companies they regulated - which is part of what causes regulatory capture over time.
There seems to be a lack of appreciation for existing legal frameworks. There are laws and precedents that have been built up over time to cover many aspects of harm that may be caused by the short run due to AI (hate speech, misinformation, and other “trust and safety issues” on the one hand, or use of AI to cause cyber attacks or physical harm). These existing legal frameworks seem ignored by many of the people calling for regulation - many of whom do not seem to have any real knowledge of what laws already exist.
Many people misunderstand that the political establishment would like nothing more than to seize more power over tech. The game in tech is often around impact and financial outcomes, while the game in DC is about power. Many who seek power would love to have a way to take over and control tech. Calling for AI regulation is creating an opening to seize broader power. As we saw with COVID policies, the “slippery slope” is real.
It is worth pausing to ask yourself who do you want setting norms for the AI-industry - the CEOs and researchers behind the main AI companies (AKA self-regulation) or unelected government officials (see e.g. current actions against crypto, and tech M&A). Which will lead to a worse outcome for AI and society?

A good question to ask yourself about regulation given recent times may include questions like - Do you think the current regulators have dealt well recently with inflation, interest rates, COVID policy, drug regulation and speed of developing life saving drugs, local policies on crime and drug use, or other issues? What primary data did you use to decide if these outcomes were good or bad or avoidable or not? This may be a litmus test for many other aspects of ones viewpoints.

Short term policy & what should be regulated?

There are some areas that seem reasonable to regulate for AI in the short run - but these should be highly targeted to pre-existing policies. For example:

Export controls. There are some things that make sense to regulate for AI now - for example the export of advanced semiconductor technology manufacturing has been, and should continue to have export controls.
Incident reporting. Open Phil has a good position excerpted here similar to incident reporting requirements in other industries (e.g. aviation) or to data breach reporting requirements, and similar to some vulnerability disclosure regimes. Many incidents wouldn’t need to be reported publicly, but could be kept confidential within a regulatory body. The goal of this is to allow regulators and perhaps others to track certain kinds of harms and close-calls from AI systems, to keep track of where the dangers are and rapidly evolve mitigation mechanisms.”
Other areas. There may be other areas that may make sense over time. Many of the areas people express strong concern for (misinformation, bias etc) have long standing legal and regulatory structures in place already.

It should be noted that I do think the short term risks of AI are overstated, long term risks may be understated. I think we should regulate AI at the moment in time where we think it represent actual existential risk, versus “more of the same” for what humanity has done in the past with or without technologies.

The first “AI election”

The 2024 presidential election may end up being our first “AI election” - in that many new generative AI technologies will likely be used at mass scale for the first time in election. Examples of use may include things like:

Personalized text to speech for large scale, personalized robo-dialing in the natural sounding voice of the candidates
Large scale content generation and farming for social media and other targeting
Deep fakes of video or other content

Whether AI actually ends up impacting the election or not, it may still be blamed for whatever outcome happens similar to social networks being blamed for the 2016 outcome. The election may be used by groups as an excuse to regulate AI.

To sum it all up…

Regulation tends to squeeze much of the innovation and optimism out of an industry. It is no mistake that many companies stop innovating when the baleful eye of a regulator settles upon them. Examples from healthcare and biopharma, nuclear, and crypto all suggest regulation can stop or slow innovation significantly, cause offshoring of industries, and derail the positive purpose of an industry. Given the huge potential of AI for healthcare, education, and other basic areas of global equity, it is better to hold off on regulating AI for now.

MY BOOK
You can order the High Growth Handbook here. Or read it online for free.

OTHER POSTS

Firesides & Podcasts

Markets:

Startup life

A brief guide to startup pivots

4 Ways Startups Fail

Founder Investors and Scout Programs

Co-Founders

Raising Money

Old Crypto Stuff:

Early days of AI (and AI Hype Cycle)

Elad Gil — Mon, 21 Aug 2023 17:59:09 GMT

I worked on early ML systems and products at Google and later at Twitter (after they bought my company, Mixer Labs). I then spent a decade working as a founder and executive & investing in machine learning companies. Until the rise of new AI architectures (in particular transformer-based and diffusion-model based approaches), roughly all machine learning startups failed. Value in prior AI waves went largely to incumbents over startups - as the capabilities were not advanced enough to create new market openings.

Here is a slide I used to use (borrowed from Brandon Ballinger) during 2017-2019 or so - this slide reflected the CNN/RNN/GAN world of the prior ML wave.

When many business people talk about “AI” today, they treat it as a continuum with past capabilities of the CNN/RNN/GAN world. In reality it is a step function in new capabilities and products enabled, and marks the dawn of a new era of tech.

It is almost like cars existed, and someone invented an airplane and said “an airplane is just another kind of car - but with wings” - instead of mentioning all the new use cases and impact to travel, logistics, defense, and other areas. The era of aviation would have kicked off, not the “era of even faster cars”.

(We should of course, fully recognize how important prior waves of ML and deep learning were and are to all this - however, to treat it as an ongoing continuum may miss the seismic nature of this technology shift).

Slide I now use.

The biggest inklings that something interesting was afoot came kicked with GPT-3 launching in June 2020. GPT-3 was a massive step up from GPT-2 and prior models. It was not quite good enough to do all the things we now view as hallmarks of “AI”, but it was highly suggestive of what was to come (I went on the A16Z podcast a few months later to talk about GPT-3, as it was so striking). For those in the know, the launch of GPT-3.5 in March 2022 solidified the perception of transformer-based models as the future. Internally at companies like Google, OpenAI, Microsoft, and Anthropic, early access to models gave a subset of people a glimpse of the future that was coming. This led to a Google engineer eventually proclaiming an internal AI chatbot named LaMDA as being “sentient” - this chatbot was a sort of predecessor to chatGPT and products like Character.AI.

The real starting guns for this AI wave in terms of a large number of founders jumping in was driven by two sets of launches. First were the launches of image-gen products like Midjourney and Stable Diffusion, followed a few months later by ChatGPT, which wowed the world, captured the public imagination, and was the AI startup big bang moment. ChatGPT truly highlighted the capabilities of these new forms of AI and the power of RLHF. OpenAI followed up with GPT-4, 4 months later.

True enterprise adoption is still many quarters/years away

ChatGPT’s launch was the starting gun for mainstreaming that AI is a big deal in terms of new capabilities and kicked off the large scale enthusiasm, hype cycle, and adoption for generative AI. This launch was only 8-9 months ago, and GPT-4 did not come out until 5 months ago. Given that large enterprise planning cycles often take 3-6 months, and then prototyping and building will take a year for a large company, we are still very far away from peak AI usage or peak AI hype. Most large enterprises are still trying to analytically sort what “AI” means for them, and are still many quarters from embracing this new technology.

4 Waves of AI Adoption

Indeed, there are likely at least 4 waves of AI to consider in these early days.

Wave 1: GenAI native companies. ChatGPT, Midjourney, Character.AI, Stable Diffusion, Github copilot, and other early launches that have now gained significant revenue and user traction. Obviously there are some great ML companies that pre-date GenAI that continue to participate in the current era (Hugging Face, Runway, Scale, WandB are a few that come to mind).

Wave 2 (current wave): Early startup adopters and fast mid-market incumbents. This is the first wave of startups to launch on top of GPT-3.5/4 like Perplexity, Langchain, Harvey or others. In parallel, a small number of founder led multi-billion companies like Navan, Notion, Quora, Replit, and Zapier launched AI-powered products quickly and are the early adopters of the wave. Microsoft, Adobe, and Google are notable outliers as very large enterprise moving fast to AI - Microsoft likely due to its inside track with OpenAI, and Adobe as diffusion models tend to be cheaper and simpler to train than the large scale LLMs.

Wave 3 (coming soon): Next wave of startups currently being founded. It will be exciting to see what is in this mix and may include new formats like voice and video in addition to using natural language in more verticals and more ways, as well as new types of infrastructure. Companies like Eleven Labs/LMNT/LFG Labs, Braintrust, and many more will provide incremental experiences. There is a big wave of new startups coming. The current YC batch alone appears to have a 100 or more AI startups….

Wave 4 (coming 2024/2025?): First big enterprise adopters. Since enterprise planning and build cycles are so long, anticipate the first really products (versus demos or prototypes) from larger companies other than MSFT, Adobe, Google, Meta to start to show up in a year or two. This is when revenue to AI infra companies will start to ramp significantly relative to today, when hype will peak, and we will see further accelerated investment in AI.

The Future Is Bright

There is enormous potential for this new wave of tech to impact humanity. For example, Google’s MedPaLM2 model outperforms human physicians to such a strong degree that having medical experts RLHF the model makes it worse (!).

Given the strong potential it will be exciting to see all the immense innovation in education, healthcare, enterprise and consumer software, and other aspects of life coming via this tech breakthrough.

We are only 8-9 months since chatGPT woke the world to this new era of AI, and exciting times are ahead as we follow this tech discontinuity down our timeline. It is the very earliest days of AI, and both peak hype and peak impact are still in the future. Lots more is still to come.

MY BOOK
You can order the High Growth Handbook here. Or read it online for free.

OTHER POSTS

Firesides & Podcasts

Markets:

Startup life

A brief guide to startup pivots

4 Ways Startups Fail

Founder Investors and Scout Programs

Co-Founders

Raising Money

Old Crypto Stuff:

AI Dev Tools Panel - Stripe AI day

Elad Gil — Fri, 07 Jul 2023 12:48:28 GMT

I will be moderating a panel as part of Stripe AI Day. See full agenda below but speakers overall include founders of Anthropic, Ramp, Runway, Zapier, LlamaIndex, Baseten, Chroma, and others.

Stripe AI day is in person Thursday, July 13th, 3-8pm.

Sign up: https://lu.ma/k1dh24e1

Transcript & Video: Claire Hughes Johnson Fireside Chat on Scaling People

Elad Gil — Fri, 26 May 2023 03:51:03 GMT

Video:

Transcript:

ELAD GIL

Yeah, I'll add two more sections about you. So we got to keep pumping sales. So I think we heard about Claire's amazing background, both in terms of Google, in terms of Stripe. You're also on multiple boards, which you may want to talk about a little bit. HubSpot, the Atlantic Aurora, the self driving car company you wrote Scaling People, which I think is sort of an instant classic. And it has a lot of different frameworks you can use as you sort of work through things. And we can talk about that a bit. But I was hoping to maybe start a little bit with your time at Google. And I feel like there's a handful of companies that at any given moment in Silicon Valley are very formative, both in terms of the people who work there, the culture that gets set and sort of spread throughout the Valley. And Google is one of those, and stripe is one of those. And so I was just hoping to start a little bit with, like, how'd you end up at Google, what made you decide to go there? And then I had some specific questions on that.

CLAIRE HUGHES JOHNSON

Yeah. And a lot you and I work together at Google, so you might also answer this question. So in what people ask me about career decisions I made, I think a few times in my life I made decisions not to do the thing everyone else was doing. And one of them was I quit my job in consulting in New York City, and my then boyfriend, now husband, and I decided to move to California and just started new life here and figured we'd find jobs. This was right after the.com bust in the early 2000s. So our families were like, don't love this plan. You irresponsible young people. And I was interviewing at a bunch of different companies, but not all tech, because that wasn't exactly my background, though I had worked my consulting firm had a technology arm, so I worked with some engineers. But I loved Google as a product. And I got introduced into Cheryl Sandberg's org by a friend of mine from business school and was like, well, would they really need me? Because it was early. I was like, I mean, I wasn't and they were they convinced me that they might need my skills. And I walked in. By the way, it was the biggest place I'd ever worked. I think it was around 1800 people. It was extremely chaotic and really fun. We worked so much because we were growing. I think we more than doubled in size my first three or four years. So from like about 2000 to 4000, 4000 to 8000, we were on a really steep and we interviewed you remember this, we were interviewing people like 40 hours a week and then doing our jobs those first initial years. But you're right, it's formative for me, certainly. But I think a lot of what Google put into practice and how it ran itself was studied by other companies. And I think some of those things translated well and some didn't, which you probably have some opinions about.

ELAD GIL

Yeah, no, I mean, I joined right around the same time you did. And I think inside of technology in Silicon Valley, it was a clear choice. Right. It was a company that was sort of sucking up all the talent, and very few people were hiring for a couple of years before that. So all the founders ended up at Google because their companies failed during the.com bus and they joined. When my wife joined, her parents actually told her it was the worst decision of her life. And then later, when she left Google to join Stripe, her parents told her that was the worst decision of her life. And so it's interesting to see how.

CLAIRE HUGHES JOHNSON

There'S that I think there's a trend. We did not listen to Jen's parents.

ELAD GIL

Yeah, I think that people, when they join companies, they sometimes pick up very good habits there and sometimes pick up very bad habits there. What do you think are the good habits and bad habits that you picked up at Google that you then sort of reset when you came to Stripe?

CLAIRE HUGHES JOHNSON

I mean, a guy that I used to work with at Google who then left and worked at another tech company and now is an investor. But he said to me that after he left, he felt like learning about business at Google, was like learning about gravity on the moon, just because AdWords was just like a money printing machine. And the margins were so beautiful. I jokingly refer to them as drug dealer margins, but, like, really unrealistic. And so I think there's some things that you just take for granted when you're in that kind of a growth with that kind of a profile of your anchor product that you have to almost, like, need to go back to basics and not think about unit economics again. The other thing is, Google, especially early, did not value management very highly. So I was sort of countercultural at Google because I always did. And I think a thing that I took away from that experience was to continue valuing it and to bring it to stripe more strongly than I felt. It was embedded in early Google, trying to think what other but there was also we had fun. I think Google did a nice job of recognition and feeling those earlier years. You really felt like you were part of something. The mission really mattered and didn't take itself too seriously. Took time to celebrate, but people worked very, very hard because you felt like you had this moment and you knew it was going to go away, though. It's been pretty strong.

ELAD GIL

Yeah, I think a lot of people forget that, because now Google is almost viewed as the opposite in terms of a lack of culture and all this stuff. But I think at the time that both of us was there, it was extremely hard charging. It was large groups of very young people, and people you'd be there at midnight, and there'd be all these people just like in the office, working and playing smart.

CLAIRE HUGHES JOHNSON

Very driven life.

ELAD GIL

Yeah.

CLAIRE HUGHES JOHNSON

Every meeting was especially in the product, you'd be hanging out aggressive, aggressively getting to the answers.

ELAD GIL

Yeah. And it was a very blunt culture, too, at the time.

CLAIRE HUGHES JOHNSON

Yeah, it was very blunt, which now everyone's like, oh, it's too nice. And I'm like, well, it wasn't that way.

ELAD GIL

Yeah, it really shifted. It's kind of interesting, because I feel like in that period of any company when there's fast growth, the founders are hyper engaged. The team is very driven. Often the founding teams decide to do almost, like, odd things. And so I remember one of the things that Larry did is you removed the entire middle management layer. He said, we don't need middle managers. And suddenly every director had 50 to 100 reports, and you saw your manager once a year because they were just too busy with 100 reports or whatever. And so a lot of people got kind of lost in that. What was the weirdest thing you experienced at Google, or what was the most unexpected thing from either managerial perspective or cultural perspective?

CLAIRE HUGHES JOHNSON

Yeah, I mean, gosh, there's so many. But that little vignette just reminded me I'll never forget I was working with the Gmail team, and I was very embedded in a good way with the engineers in particular, because we had just launched it. There was a lot of I was on the operations side, but Google had not really had a product that required as much support and explaining because Gmail was a little bit of a new paradigm. Users were very lost. Anyway, I was embedded with the engineers, so I would hang out near where they all sat. And after this middle management layer thing, one of the engineering, I guess, directors then made, like, yelled out into the cubicles, hey, everybody, your bonus letters are in a file on my desk. Come and take yours. Because he didn't have time to do one on ones and hand them and talk to them about their compensation, and I hope they were in individual envelopes, but I'm pretty sure they weren't. So everybody was just rifling through each other's bonuses amounts and taking their letter to find out how much their bonus was. And I was like, this can't be right. This cannot be good. Anyway, a lost opportunity for that. Maybe, but maybe Larry was right. My first meeting with Larry and Sergey actually was about Gmail. We were pushing really hard to get the product into 60 languages. That was the whole push for the whole company, was to get every product into 60 languages within like 30 or 60 days, depending on who you asked. And I was supposed to scale that support of that product in the 60 languages. I had a team that spoke about eleven languages, which was impressive, by the way, because the team was only about 18 people. And so, yes, you can imagine what happened to the folks. The one Korean speaker, for example. So I wrote this pretty extensive presentation on what it was going to take. And we went into Larry and Sergey in one of the conference rooms where they used to have a lot of meetings. And I walked through not much of it because they're pretty impatient and don't tolerate. I was not being very MBA consulting, but I probably felt very much that way to them.

ELAD GIL

They would also exercise in the middle meetings.

CLAIRE HUGHES JOHNSON

Oh yeah, there was StairMaster, and Sergey would be sweating, and it would smell like a locker room, and you're like, trying to explain anyway, that was a different meeting. But after a minute, they looked at me and they were like, no, this is a bad plan, but look at the analysis. Like, look at these numbers. No, do something different. And I was like, yeah, great. Okay. But it was kind of great, right? I think there's like, constraints do create innovation. I ended up coming out of that meeting and building the first peer to peer support platform for a Google product because I was like, well, if we're not going to answer your questions, someone else is going to, and we're going to build a way to do it. But at the time it was demoralizing and embarrassing and all the things, but I now kind of love it as one of my so was I shocked? It was shocking in the moment, but it was kind of amazing.

ELAD GIL

Yeah, they really forced people to think big and do really radical things. One thing that got sparked off of your comment on the bonuses, because I don't know if you remember this, but they used to give out $1,000 in cash to every employee for Christmas.

CLAIRE HUGHES JOHNSON

Yes.

ELAD GIL

And it'd be ten hundred dollars bills. And there's this whole debate internally of like, should you give it in cash? Should you give a check? And they wanted it in cash, so you'd viscerally feel the money and spend it for Christmas, for the holidays. And I think eventually there used to be like the rumor is there'd be people who'd wait for the Google buses and SF so that they could mug people coming off with $1,000 checks, like every holiday season. And so then they had to discontinue.

CLAIRE HUGHES JOHNSON

It, but it would be at the holiday party and the Brinks trucks because we got bigger, and they would pull up and there'd be all these armed carts at the end of the holiday party because we were all walking out with $1,000. I will tell a story about that, which is at the time. So I joined Google, got married, had my daughter, and then had my son. It was like a pretty intense period of my life. And I remember telling a younger employee that I had taken my $1,000. My kids were in the Google Daycare. But by the point that I had my second child, we'd hired this part time nanny who would help. My husband was working full time, and there was no way I could have done all the things without this help. So Berta, who was with us part time for a long time and is the reason my son has a beautiful Mexican accent and speaks Spanish beautifully. Anyway, she would speak to him constantly, but I would just give her the $1,000. And I told this to someone. They were like what? You're giving your bonus? I said someday you'll understand. Like, that woman is helping me raise my kids. Of course I give her my bonus. Like, it's her bonus. But later, that person, that employee came up to me. She's like, I totally get it. She's like, I get it now. But if you're a working mom, especially, you need help. Especially if your family is far away, which mine was. So Bertha was our family.

ELAD GIL

Yeah. It's interesting, the point on innovation at Google and how it forced a lot of people to do very innovative things. Like you building the first peer to peer sort of support platform for the company. Why do companies slow down as they get bigger?

CLAIRE HUGHES JOHNSON

I mean, you have a lot of thoughts on this. I want your opinion. Yeah, I think that there are two things that come to my mind. So the first one, if sort of my book does any trick for people if you don't build in two things, one is smart. Not heavy processes, not painful processes, but smart ways of operating that get the work done in a repeatable. Like you should be able to repeatably do important things for your company to keep velocity right. And if you don't figure out a way to scale, like product launches or marketing campaigns or sales processes, you will start to fall over because you'll be manually building this stuff, and you'll have so many interface points and custom elements that you can't scale it. Right. So that's one reason. The bigger reason, I think, is that founding teams, including the executives and the leaders don't figure out how to push and delegate down decision making and empower and build confidence in the people they're trying to hold accountable. So you'll get this weird phenomenon where people are like, well, I'm accountable for this product, but then yet they're waiting for some meeting to happen to feel like they've got enough alignment and consensus, often with the founders to actually make the controversial decision to further the product. So they're accountable, but they're not actually owning it. And I think that is very poisonous. And I think the other, the third would be when you have a moment, as Google has and Stripe has, I hope continues, where you have a real opportunity in a really important space where there's a lot happening with the technology you're building and it feels ephemeral. You try to do all the things and you don't prioritize, and then you get a bunch of smart people trying to read your priorities and you haven't been clear enough about it. Right? And I think I remember Eric Schmidt when he was CEO going on some road shows because I think he was realizing, nobody, we don't all have the same priorities and it's a problem. And I think we went through some similar things and have gone at Stripe, but those are some of my thoughts. But it's really about not recognizing that the stuff's falling over and then changing leadership behavior.

ELAD GIL

What specific leadership behavior would you change?

CLAIRE HUGHES JOHNSON

This is the hardest thing to say to a founder, which is, there are things you care about that you cannot be in the room anymore. You will not be the decision maker. And I think the best thing you can do is say, what can we do to help take that thing out of your head? I'm like Picturing in Harry Potter when they take the thing and they put it in the pool of water, like that little thread of judgment, right? And belief, like, this is my vision and this is why this has to be this way, and this is my quality level. How can you imprint that on others? I did a first round interview many years ago where I thought, I think the medical school, c one, do one, teach one, come up with patterns that you as a founder can use to gain confidence that you can trust the people that are going to need to run without you. And unfortunately, I think part of the nature of being a founder is you're innately a bit paranoid and you don't have that kind of trust because that thing is your baby. Like, you created it and you want to protect it and you want it to flourish. And you certainly don't want the thing. That's the irony of this. You don't want things to slow down. You have so much ambition. You want to go faster, but to go faster, you have to give up a lot. And also let people fail and it's hard to watch people injured the thing that you've just spent the last five years building. So I think that's the main behavior is delegating decision making and accountability, teaching owners, teaching the values so that you don't have to be in the room.

ELAD GIL

One of the arguments that I know a founder of a very well known public financial services company, like one of the biggest in the world, and her view was that it's really hard to be both big and innovative. And the reasoning was sort of threefold. Number one is you shift from building mode to compliance mode in certain ways, particularly in regulated industries, right, in certain industries. Second is you kind of start building processes to help the average person do well versus the best people do well. Right. And so you have sort of reversion to the mean and that slows things down pretty dramatically. And then third, the nature of the people that you hire shifts because as you get bigger and become a bigger brand, you start attracting a different type of person. The people are less entrepreneurial.

CLAIRE HUGHES JOHNSON

Well, and they're more specializing in doing their particular local thing well, as opposed to optimizing for the global. That's true.

ELAD GIL

So how do you break that pattern? Because it seems like it's very consistent. And the question is, is that a universal pattern? Is it breakable like other ways to change that?

CLAIRE HUGHES JOHNSON

I mean, I think the things we've certainly tried some of these that even Jack Welch talked about this, which is you hive off parts of the company and you protect them. I mean, Google did this with Gmail initially as they put them in a part of a building that no one else had access to and sort of separated them even from young Google to go and build on their own and experiment and launch. But I think one tactic is for something that is innovative or new. So this is on the innovation point, you eventually do have to separate that part of the DNA out and protect it and treat it a little bit differently, which doesn't, to your point, things regret everyone wants. Well, what about fairness? What about fairness? I'm like, well, it's not actually fair. You're going to pick some people to work on something and you're going to give them special permission to do some things differently. Right? I think that's what sort of Google X tried to be and it's hard, but that to me seems to be the most prevailing pattern to fight that. I think on the regression to the mean and the specialization versus the generalists, some of that is good. I mean, that is how you get sort of I think if you think about McKinsey's sort of growth curves, if you're talking about your anchor mature product, that's very normal to start to solve for more of a mass, like whoever's going to do this onboarding as a customer, those customers are going to be more diverse. They're going to need more support. It's going to need more it's not a regression. It's more of getting to an average, because that is what success is. And having specialists who understand compliance is you're now getting regulated. There are people scrutinizing you. You do not want to get in trouble, certainly not on behalf of your customers. I think that's normal. I think the problem is the next waves that you're working on cannot have the same thing happening. And to finish my own, you can carve off a small team and treat them differently, but can you treat a whole additional business unit differently? Because you're thinking about this growth horizon five years from now. And that one I don't know of a lot of companies that have done that well, I don't know if you have, but I think it comes I mean, often my answers will come down to leadership. And I'll be honest, this is an imperfect example, but when Larry made the decision to buy Danger and bring Andy Rubin in, he basically did this at a bigger scale, which is, I'm going to take this thing. I'm going to bring in a leader who's more entrepreneurial and innovative, and I'm going to give him a big portion of resources and let them do what they want to do. And it was controversial internally, right? People really fought that. But he was building for the future, and he had to separate it from the mothership.

ELAD GIL

I think the acquisition of Android in particular, too, one of the things that they did differently is they actually let Ruben divorce the hiring pipeline from the core Google hiring process. And so he hired all sorts of specialists who never would have made it through Google hiring, because at the time, Google only hired generalist engineers. And if you really, really knew how to hack a Linux kernel, it didn't matter even because you were sort of gauged on all this other stuff. And so I think that was one of the biggest changes that actually allowed them to scale.

CLAIRE HUGHES JOHNSON

Yeah. So I guess the answer is, I think it's very hard to do it all in one unit. And so how do you separate it and how do you hire the right leader for a very different enterprise than the one that is sucking most of your attention and energy, which is your core product?

ELAD GIL

How did you learn about Stripe initially and what made you decide to make the leap?

CLAIRE HUGHES JOHNSON

So Stripe did a very good job early on, actually, of getting attention, what I would call it, like sort of earned free media from I mean, Patrick and John are fantastic spokespeople. They hired a comms leader early who I think did some very smart work getting the word out about the company. We didn't really have much of a marketing team. We had one product marketer. It was not our own marketing effort. It was much more of a media effort, but also very present in Quora and Hacker News, and IRC, a lot of channels where at the time, developers were really our main audience. So there was buzz about stripe before it was really even a thing, in a way. And I think it's because developers appreciated the product and they were talking, and I was close enough to enough engineers at that point that I was hearing about it. And then I feel like I read one profile in a more mainstream, maybe Wired, maybe something like that, and I thought, you know, I admire what I'm hearing from these founders. So it was like in my brain, but barely. And then actually, a former colleague of mine at Google who'd gone to work at Sequoia contacted me and said, I'd really like you to meet Patrick Carlson, and they think a COO, and I think you should talk to him. And I said, no. I was like, oh, payments. No. He said, Claire, have I ever wasted your time? Of course, he's my friend. So I was like, all right, I'll have breakfast with the guy, whatever. So I meet Patrick and found it was a compelling meeting. But we both, I think, had to go back and think, like, he wasn't sure did they want a COO ahead of sales. He was meeting a lot of people. And then I had just taken on a new role in self driving cars at Google, and I felt committed to them, and I was sort of playing through that. So it was a slow process of getting to know each other. But I do remember a moment a lot where another guy at Google, a leader that I knew who had reported to me at one point but was now a colleague and a friend, said he had met with them. And I was jealous. And it was like, this is where you had these terrible dating analogies. But I was like, wait a minute. I might care more about that opportunity than I thought. And so that was helpful. The next time Patrick got in touch, I was like, yes, let's meet again.

ELAD GIL

And it's interesting because payments went through a period in Silicon Valley where it was not considered a good place to work. And there's a few reasons for that. And I think one of the reasons.

CLAIRE HUGHES JOHNSON

Google and I worked on it at Google, and it was brutal.

ELAD GIL

Yeah, Google had kind of payments product that didn't work quite as well. And then the thinking in Silicon Valley at the time, and I think this was really spread by the PayPal group. The folks at PayPal said that payments were too hard because there was too much fraud, because when they launched, there was a lot of fraudulent behavior on PayPal. And so there was this meme in Silicon Valley for years that you should never do payments because you could never get over the fraud issue. And so.

CLAIRE HUGHES JOHNSON

That's why had to hire thousands and thousands of people in Omaha.

ELAD GIL

Yeah, you had to hire all these people and do all this stuff. And so there was like a five to ten year gap where there was just no innovation in payments.

CLAIRE HUGHES JOHNSON

Yeah. I think there was also a little bit of and Google had this challenge, which is young tech companies often don't partner very well, and you really have to partner and build into the ecosystem and be credible player early with banks right. And financial partners. And that was not a strength of Google's, certainly, and I was part of some of these experiments, but I think it also became almost a meme beyond Google, which is like, well, you're never going to get these old stodgy, established financial institutions to talk to you. You little young startups, right. And Stripe was like, yeah, we are.

ELAD GIL

Yeah. So how do you think about it from a careerist perspective? And when you think about your career, were you ever doing, like, career planning, or did you just follow what made sense in a given moment, or how.

CLAIRE HUGHES JOHNSON

Do you think I think looking back, it makes a lot of sense to me. But in the moment it was more and there's a chapter in the book called you where I talk about sort of how I pick my head up every six months or so, but really on twelve to 18 month cycles, like, am I learning? Am I in a role that I think could lead to a role? And it's never a title, it's more like, am I going to be doing stuff six months from now that I think I'm going to be learning and having an impact? And it's roughly on a trajectory that makes sense to me. And I think what I loved about my opportunity at Google is I had like, five different jobs in ten years, and they were all quite different, and I learned a ton. I mean, it was a great development journey for me. I didn't really know what it was all going to amount to. I wasn't angling for a particular direction. But it did get to be a point toward the end where I was a very, I think, viable COO candidate. And I was talking to a lot of growth stage startups, and I actually had a very refreshing conversation with Reed Hoffman at one point where I just felt like, look, I've got this job, I've got kids, I can't meet all these people. And he was like, Claire, there's only a few companies at any given time that you should really be considering, and you know who they are. I was like, oh, he's right. There's not some secret thing going. I mean, I'm in Silicon Valley, I have a lot of network, I know a lot of investors. So I started to realize I could be more targeted with paying attention to what was out there. But the real decision came down to. I mean, one for me, I was definitely not learning at the same rate at Google. And even though I'd moved myself into this very entrepreneurial and new and like almost an R and D project with self driving cars, I was pushing up against some stuff that I was like, I'm never going to sort of win this internally. Like, I have made proposals. They were getting rejected. I thought we should have been swinging for the fences. We were out ahead with that technology. A lot is nodding, and now there's a lot of others in the race. Waymo is still there. I actually really admire the co CEOs of Waymo. They're friends of mine, both of them. So I just want to be clear about that. But we're talking now about almost ten years ago, right? Anyway, point is, that was frustrating and I didn't feel I was learning. And I felt like, what am I going to regret? So you get to a point in your career where you're like, you've done a lot, I've had some success. What am I going to regret? And I realized I was going to regret not going earlier into a company and thinking, could I build? Can I help build? Because I joined Google as a manager, right? I eventually left as like a VP, two, whatever, but I really was more rising up through the ranks of Google, which is, by the way, where you should look for talent, pick a great company, someone who's had a great trajectory, which I did there, and reputation and references. But I think I needed to see if I could do it myself, like, from a more executive position. And then stripe represented so much of what I realized I was looking for, which was not payments, I admit, at least not initially, but though ultimately very strategically interesting. It was one, the founders, like, they really matter, and they have to be excellent. It's like you're investing it, but it's your time, not your money, right? The other was I thought that the company was meaningful. Like it had a mission, it was infrastructure, it mattered, right? I was sort of tired of this ad based revenue play. Like, I really wanted to do something that felt important. And I think actually, economic infrastructure is very geopolitically important. I'm more of a mission driven kind of person. Economic growth matters to me at a macro level, and then I could see that I could have an impact. It was clear that they were getting to a point where they could use my skills and I could learn new things, and they were going to give me some stuff to run that I'd never run before, which admittedly, was a trust exercise. And in the end, I called my husband one night, I was like, oh, if I don't take this, I'm never leaving Google. And I knew I was leaving. And so I was like, well, that's obvious then, but it took time to get there. I think there's some lore around my interview process, but it was long, so a lot of mutual diligence. And in retrospect, actually, I felt like it was a big career risk for me. Right. I was, like, really thinking through it. But it's a bigger deal for the founders. It's a bigger deal when you're bringing in, especially COO has a little bit of baggage, good and bad, attached to it. Even if you don't conceive of the role as some kind of proxy, it's conceived of as a proxy, and you really have to be careful. Like, is that person going to be culturally appropriate and good? But both of those things because you can have a real impact on a young company in that role.

ELAD GIL

Yeah. Your book talks a lot about different aspects of leadership and management and mentorship, and I have sort of two related questions. One is, when you come into a situation like that, how should you as a new executive act or what are the first things that you should do?

CLAIRE HUGHES JOHNSON

I mean, I felt very educated coming in because we had spent a lot of time together. I'd looked at a lot of board presentations, internal documents. I looked at the company goals, which I helped write as part of an exercise for one of my interviews. And I knew Patrick and John and Billy, who was a leader in the company quite well at that point. So that was an advantage because the first thing you got to do is let's build some relationships. Let's talk about what do you do well, what do I do well? How do we make decisions together? Is this a divide and conquer on some things? Is this a joint decision making? What kind of leadership team are we building here? But really, I listened. I mean, I did. You can read the first 90 days, but a lot of that stuff is legitimately reasonable, which is, of course, you're getting pulled in immediately to other things. But I worked to learn the product. I worked to meet everyone in the company because it was small enough that I could do that. I took a ton of notes. I distilled it all down to what I was hearing. Just like any good consultant, I was like, here's what I'm hearing. The people internally want to see us do differently or that we need to tackle or we need to build. And I played that back to the whole company, which was well received because everyone cared a lot. They were building it together, and I think they were worried I was going to come in and one, take Google and try to imprint it on stripe, which would have been really inappropriate, by the way. Look at the products, look at the business model, look at the space in the market. All of that was different. And so I think you have to sort of rip. There's a little bit of brain, like, dump all of that and start over and be curious and learn and listen, but then start to have opinion. After about a month or so, you've got to have some opinions about your priorities and what kind of thing you need to do. And honestly, I've said this in a few forums, but I think bringing in some talent if you're in a growth stager, we really didn't have a lot of management experience and starting to think about, okay, if we're going to scale international on sales. For me, I was like, wow, we need to rebuild some of the support team. So starting to do a lot of recruiting. And if you see a new leader and they're not starting to think about talent, they know and holes you might have and putting those things together, I think that's not a good sign. Right. But yeah, it's forgetting the things that your old constructs and building new constructs, but then being smart about where your opportunity is in the near term. You also want to have some impact pretty early. What are you positively viewed?

ELAD GIL

Yeah. What do you think are other signals that a manager isn't necessarily going to scale? Like, I think to your point, one example is there's no sort of team roadmap.

CLAIRE HUGHES JOHNSON

Yeah. They're not attracting talent, they're not thinking about where the team needs to be. I think that's one of the signs. I think another one is I have too many founders come to me and they've hired someone with experience, and then they're kind of puzzling over, why does that person with experience seem to keep checking with me? Right. And I think it's kind of this problem, which is a little bit of an uncanny valley where it's like, you don't want to come in and disrupt the Apple Card and have organ rejection against you as a new leader. So it could be that. But often, unfortunately, I think it's not a great hire because they're not coming in with a sort of confidence from their experience and thinking, okay, well, two years from now, if we're going to be as good at sales as you seem to think we need to be, then we're going to have to do some things differently. And instead they're sort of check checking and the founder gets like, this feeling in their gut, like, OOH, this might not be the person, they might not actually see the field the way I need them to. You're probably right. So I think it's slow to move, slow to hire, too much checking, no opinions.

ELAD GIL

Yeah. When you give people advice on what to do career wise, and again, I think there's really good nuggets in your book. What is the difference in the advice that you give to somebody who's early in their career versus late or later?

CLAIRE HUGHES JOHNSON

Yeah. If you're early in your career, well, one, don't do the thing everyone else is doing which unfortunately is like a siren song to young people. And I get it. But I taught a class at Harvard Business School last fall and so many students came to my office hours and were like, I'm going to go work in venture capital. And I had to spend the entire office hour being like, that is the worst idea I've ever heard. No, but I mean, I was like, go do something. Are you kidding me? That will always be there. And what are they going to do? They're going to be like a spreadsheet monkey. Like anyway, sorry but don't get me started on this, but I really think introspecting check, gut check. Like, okay, a lot of people around me want to do this thing. Maybe that's right for you, but it's probably not. It's that no one can think of a better thing and you've got to know yourself. And what are you curious about? What are you interested in? You can take so much more risk when you're young. I mean, I did a ton of different random stuff in my 20s. Like really random. I mean my career in Tech didn't start till I was in my thirty s and I wouldn't trade any of that experience. I worked in politics, I worked in journalism, I worked in consulting. I did a bunch of different stuff. I made very little money, but I was fortunate. I didn't have a ton of student debt. I could live at home, I could do some things to save up. But I think that one follow your gut, not other people's gut. Be learning constantly, asking for feedback, but not in the grubbing for feedback way, but in the like, really, I want I'm hungry. And I think the other thing is I worked really hard. I mean, be ready to lay it out there and demonstrate and then people notice you and then people introduce you. I'm still surprised at some of the people I worked with in my early career who matter to me now in my network, who are in a totally different world than Tech, but they still one of them is a former governor of Massachusetts who I keep running into now that I live back there. And he's amazing and he definitely remembers me and talks about me to other people, right? Like that matters. He was not the governor when I knew him. Just be clear. So those are some thoughts. I think later in your career you have a different often risk appetite, which is unfortunate, but I think you kind of have studied yourself enough. I hope you're more self aware about what you're great at, what you're passionate about, where you can add value and that you can be more discerning about where you should play or maybe you want to try something new. But I really think that too many people, again, even in their later career, are following the hotness. Right? You don't want to go work at the company that's hot right now. You want to go work at the place that's going to be and for some reason risk aversion takes over. I don't know. Be smart, do the analysis yourself. Don't listen to what other people think. Maybe that's my theme of this whole answer. What do you think?

ELAD GIL

I give slightly different advice. I think that following your passion sometimes is good but sometimes is really bad and it depends on what your passion is.

CLAIRE HUGHES JOHNSON

That's true? Yes.

ELAD GIL

And so I would really qualify like where your gut wants to take you versus where reality like if you're a.

CLAIRE HUGHES JOHNSON

Terrible poet and your passion for writing poetry but perhaps will not be a good career.

ELAD GIL

Exactly. And I think a lot of people kind of follow that version or follow.

CLAIRE HUGHES JOHNSON

Your curiosity in a direction that you macro environment you're like that analytically makes sense and emotionally that makes a lot of sense.

ELAD GIL

Yeah, and then I think the main thing is there's the old saying that if you see like a rocket ship taking off, get on it, it doesn't matter what seat you have. And so I think a lot of people will start debating role, they'll start debating comp, they'll start debating all these things instead of just saying like, this is a great company and it's going to grow really fast and it's going to do interesting things and I should just get on it 100%. And usually those things are messy and they're ugly and they're painful and you.

CLAIRE HUGHES JOHNSON

Might be in the wrong role and.

ELAD GIL

You have to work really hard. Yeah, it all works itself out.

CLAIRE HUGHES JOHNSON

I agree. It's like take the best professor in college, not what the course is about. I think it's the same thing. Best company wins.

ELAD GIL

Yeah. So I think early on that's the biggest optimization function later in the career. I think it's a little bit about creating doors because I feel like too often people just it's your point on following a specific path, but I also think they stop creating optionality. And I feel like some of the best career moves are like when you're like, oh, I worked on ads my entire life, but I'm going to go to this company that's doing something completely different and have new things. I'm focused on functionally and new things because you kind of reach the peak of what you were doing before.

CLAIRE HUGHES JOHNSON

Yeah, I don't think I actually have this in the book. I always say to people I've always had two and sometimes three potential next paths that I felt were available to me, like career wise, job wise, because you never know what's going to happen, right? This door could get closed. Something could change in the environment in you want to pivot. Like I always had a theory about, okay, well if that happened, if this you kind of want to think a few steps ahead. Whatever organization you're in or political system or whatever it is, and just know that you have another path. I agree. I think some people get pigeonholed and trapped. That's the other thing. People just are afraid to change.

ELAD GIL

Yeah, they get stuck at some point. They get stuck. So we talked a little about how Larry and Sergey ran things in a very different way at Google and everything from, like, getting on the StairMaster in the middle of a meeting to really pushing people to think big. And I think that repetition on thinking big, for example, translated I remember when I first started working on mobile products at Google, I met with, I think, Paul Buche, who started Gmail.

CLAIRE HUGHES JOHNSON

Yes.

ELAD GIL

And I talked about, oh, wouldn't it be great if we had Gmail on mobile? Because I was working on mobile and it was a new thing. And he's like, well, how it'll be different from all the other mobile apps and how it'll actually be interesting. And a big push on why is this different, which I think was really sort of Larry and Sergey thinking, like, translated through the team. What do you think was a version of that at Stripe when you joined in terms of either the quirkiness or the maniacal focus on something?

CLAIRE HUGHES JOHNSON

I think that one of the things that I heard about Stripe when I was looking at it was that there's a little bit of a trend that sort of has turned. But I can think of some products I use right now that hasn't, which is that B to B products were not well designed. They were not, like, pleasurable to use either as a developer integrating them or as a user. I'm talking about UI design, simplicity, intelligence, and that Stripe had counteracted that. It was like, no, this is definitely a B to B thing, but we care about craftsmanship and design and experience, but are also quirky and have personality. But I really think that that care for the details and the quality. And then you'll hear a lot of people, Stripe say, we abstracted away complexity. If you think about payments and the infrastructure and the integrations and all the things involved, and all you want to do is accept payments or move money in or out. And Stripe really tried to create an elegant way to get into that system without having to handle all the headaches. And that takes a lot of deep work and craftsmanship. Any review of anything that was going to get launched was serious at Google, but way more serious at Stripe. And you're also dealing with people's money, and so it should be very serious and work very well. Right, but I think that I thought I had a sense of good design and things that were ready to be tested by customers until I got to Stripe and I was like, oh, or maybe not. And you could never launch anything if you had not tested it with a bunch of customers and were in the room, representing their feedback honestly and directly, which Google is not particularly customer oriented because, again, it's more of a consumer, mass consumer company than a B to B company. And I loved that combination of the focus on the customer, the design, the craftsmanship, and the quality.

ELAD GIL

Bar so you had an amazing career in terms of working at two iconic companies and key roles.

CLAIRE HUGHES JOHNSON

What made you decide to write a book a lot? You are an example to me, an inspiration. Now, I really think it started with you. The chapter that I contributed to in your book got a lot of more attention than I would have expected. Patrick noticed, and then Patrick and John decided I needed to write a book. So it's kind of your fault.

ELAD GIL

That chapter really took off by the way. It was circulating with a lot of different founders, and I saw a lot of people starting to write like, guides to working with themselves.

CLAIRE HUGHES JOHNSON

Yeah, you're in the book. I talk about this in the book a lot. I can't believe you said you're saying.

ELAD GIL

You have a whole chapter.

CLAIRE HUGHES JOHNSON

You made me include the example of the working with Claire guide in your book, and you said, this is going to be like catnip to founders. And I was like, really? No one wants to read that. People don't even know who I am. You were right. I mean, that's probably the thing that gets, like, tweeted at me or the most of anything that I've ever done, which is interesting. And every journalist, by the way, who interviewed me about this book wanted to talk about that because it's sort of fascinating to people as a concept. So, anyway, your chapter, I think that what John and Patrick felt, and I had to agree with them ultimately, because I did some research on my own. Was there wasn't a tactical enough, like, really practical, more like a reference manual. You look in the table of contents, I want to learn about this thing. And I'm not saying I'm the expert, but I have had about 20 years at two companies that have done well. It's more of my patterns, my examples, a lot of examples from Stripe and some from Google that I've anonymized. But I think that you want to see how someone else did the thing because it's easy to abstractly say, well, this is how you manage someone out. But in the book, I really walk through all the steps that you might take, and I think a lot of that stuff happens behind closed doors or how you put up create structures to run a company. Some of it is a little bit boring, but it's really important. And no one's written it in gory detail because it can be a little bit boring to read, but you want to see it.

ELAD GIL

I think it captured a lot of super useful stuff that nobody's captured anywhere else. And I think that's one of the really unique things about it, and I think it's interesting to see how that gap has persisted. And then you see these books that are 500 pages long with one concept, right?

CLAIRE HUGHES JOHNSON

Yeah, there's like one prevailing framework of a lot of management books, and you can get it in the first intro, chapter one. This is not that book.

ELAD GIL

That's great.

CLAIRE HUGHES JOHNSON

This is more like the encyclopedia of what's in my head. But I've been really glad that it's been useful and I want to thank you. You did inspire.

ELAD GIL

No, thanks for being part of it. What was the thing that you omitted that you wished was in there? Or is there anything like that? Or if you were to do an addendum, what would be in it?

CLAIRE HUGHES JOHNSON

I have found that as I've been doing this sort of tour and I feel very over promoted and exposed right now and doing podcasts, I have stronger opinions about some of the things in the book that I probably lightly was opinionated about. So it's not that I omitted a particular section, but I think I should have been a little more I could have even had a version of a sidebar where it's like, if you really want to know what I think, lay it out.

ELAD GIL

What's an example of that?

CLAIRE HUGHES JOHNSON

Well, lately I've been talking to a lot of I think you've got to move talent out quick, bad talent out quickly, but in particular, leaders that aren't working more quickly than I even ever was comfortable. I have so much conviction. They can just be such a rot on the culture and on the trajectory. And it's problematic because leaders are not like if they're really overtly bad and toxic, yes, of course they're gone, but I'm talking about the sort of mediocre fumbling in the darkness for six months killing you if you're a young company. That's one thing I've gotten more virulent about recently.

ELAD GIL

What was the most unexpected aspect of writing a book or the most unexpected outcome of it?

CLAIRE HUGHES JOHNSON

I mean, it's like going back to being an individual contributor. I've been managing and leading teams and not done a lot of the work myself, and I felt like the maker schedule versus the manners. I had to write a lot of this book, even though I blocked time during the week, on weekends, when I had an entire day where I could only do the thing. And I hadn't had that for a long time. And I also, I think, hadn't had as much empathy for some of the people who are trying to be makers at stripe. So one was just changing my brain and my ability to concentrate and focus and get the stuff on paper. It is important. I'm more of a talker than a writer, and so to go back to that and try to express myself in writing was like muscles. I hadn't used enough. So it was painful, to be honest. It was pretty painful. But it's very rewarding to me that some people have said to me that it's enjoyable to read and they thought it was going to be very dry and that means a lot to me because ultimately not that you all I'm the child of two teachers. Like, I grew up in a household that valued education and writing and reading highly. And so it's kind of a miracle to me that I've written a book and if someone enjoys reading it, it's very meaningful to me because that was a value system in which I was raised.

ELAD GIL

I guess I'm going to ask you two last questions and then if you have a couple more minutes, we can open up for the audience a little bit. You've had this amazing career arc. You're now doing boards, you've written a book. Is there something specific that you're hoping to accomplish over the next ten years or a specific plan or how are you thinking about what's to come?

CLAIRE HUGHES JOHNSON

I am iterating toward that I think I am learning already some things that I would have done a little differently. I would not have joined quite so many boards. Too much commitment. I think that there's some very interesting so here's my theory. My theory is that I'm going to do some portfolio of things while my kids are still at home, which is only a few more years because I enjoy flexibility that I didn't have for most of their lives. And then there are so many interesting things happening right now. I'm paying a lot of attention and trying to get more educated as again, more of an individual contributor again and hold on to that and not jump in and then we'll see. But I don't have a particular plan. I don't think I'm done right. It's just sort of like there is another chapter, I just don't know exactly what it looks like. But I'm not no longer anxious about that. I'm kind of interested and I don't feel like I've had such a I know you said to me the career, it's the impact you have on other people. It's the things I haven't done that I think more about than what have I accomplished? I find it very surprising because I think a lot of people could have done what I've done and many have and they just don't have they don't have the PR team that I have. I don't know.

ELAD GIL

So I guess the last question on my side then we can open it up is I've used Stripe as a very optimistic company in some ways as an external viewer. Yeah.

CLAIRE HUGHES JOHNSON

We're like macro optimists and micro pessimists, actually.

ELAD GIL

Yeah. Which I think is a very powerful combination and it's a very tech forward company. It has acted really early on things like climate from a corporate perspective, like it's very innovative. And I feel that people that come out of Stripe often have this sort of optimistic can. Do view of the world. What are you most optimistic about for the future? At whatever level you want to address that question.

CLAIRE HUGHES JOHNSON

There'S two things that just sprung to mind. One is an argument I had with a friend in New York a few weeks ago where he's convinced that we might not be alive in ten years because of climate change. And I was convincing him that, like, the brightest people I know are working on the problem. And I have confidence that humans can do better than this. So that's what I'm optimistic about. It feels very dark right now, but we can innovate. I know we can, and we have to. And so that then on the way here, I had this fantastic Uber driver. We had the most amazing conversation. He was asking about my book, and then I was like, no, let's not talk about that. So then we started talking about AI, and he'd been listening to Elon speak, and he's like, what do you think of this? Is the computer going to kill all of us? And I found myself not. I was like, well, this is, like, pretty freaking interesting. I guess what I'm really I'm excited that we're alive to see this a lot. Isn't that? So what am I optimistic about? Is like, it's going to be crazy, right? But you have to be a learner. You have to stay out there and stay curious and think, what could this all become? And I'm really interested. I read a lot of news. I'm on the board of the Atlantic for a reason. I really care about all the levels, like, the doers in the middle of all of this, but also the commentators. And I really am interested in all the observations around what's happening, including what's in the middle. And I feel very fortunate that I have people to talk to on all the levels, and I want to take advantage of that. I want everyone to take advantage of that. Because it's when you realize you're in history, you get old and you're like, oh, my gosh, this is history. We're in it. We are in it.

ELAD GIL

That's very exciting.

CLAIRE HUGHES JOHNSON

And I think we can impact it. A lot of people in this room can impact it. That's got to feel good, right? Scary, but good.

ELAD GIL

Yeah. So amazing note to, I think, open this up to the audience. So are there any questions? We can go in the back.

AUDIENCE QUESTION

[ASKS A QUESTION]

CLAIRE HUGHES JOHNSON

Yeah, I mean, the question was, like, some companies really fall over because they grow too quickly. They raise a bunch of money, they hire a bunch of people. They're really not ready for all that. And when I came in, I came into Stripe with about 160 people. And actually, Stripe had sort of the opposite problem, which I think we need more companies to come back to, which is they definitely had product market fit and traction. I think we're a little bit paranoid about is this real? And you could see it, I could see it in the numbers and if you looked at the inbound customer demand, both on the support side but the sales side, we couldn't get back to the leads. I did a whole analysis when I came in that was sort of bottoms up, tops down, and it should have been more like 500 people when it was 200 people. And we talked about that internally and it kind of freaked people out. But I think that there was a lot of conservatism and a desire to really have a healthy PNL and charge for the product and scale in a intentional manner. And I think that's a better problem to have being underbuilt for where you are than being overbuilt. I mean, a lot is nodding. You're creating a bunch of process for like you don't yet have. You've really got to feel a lot of conviction that you have a product that people really, really want and you can't get enough of it in their hands. And I think too many companies are like, well we got some revenue, let's build all the things. And so I think there's probably you could talk about this more, but when we say the words product market fit, different people are talking about different things. Stripe very much had it like well beyond it when I joined. And I think that if I had joined, I don't know when it was 40 people, stripe hung around, by the way, 30 to 40 people for a while, really testing and finding. Did we really have traction, was the growth really going to be real? That would have been a bad I would not have added a lot of value then in that phase. But it was a decent size, 40 people. I think it's product market fit. Let's be really clear about what we mean when we say that is probably.

ELAD GIL

Yeah, I totally agree with that. I think people misinterpreted in all sorts of ways in both directions. And I think people have forgotten how capital efficient companies used to be in part due to COVID and zero interest rates and all that. But the reality is the best companies in the world often were extremely capital efficient because a product fit was so strong that you could be capital efficient or you'd effectively be always under hiring because things would be growing faster than you could keep up.

CLAIRE HUGHES JOHNSON

Yeah, there was a joke when I joined Stripe that they kept talking about AR like annualized run rate and I was like, guys, let's just talk about the actual revenue because I think we need to graduate from that because that's just forecasting hope. I mean, Stripe, it was more repeatably because it was very transaction based, but I think there's a lot of SaaS companies that are taking something to the bank that's not in the bank.

AUDIENCE QUESTION

So, Claire, you had mentioned that your chapter in Elad's book was a genesis of this and became a good product that John Patrick had encouraged you to embark on. And Elad had mentioned that there was a corpus of knowledge that isn't really covered in other books. Claire, I'm curious, is there a corpus of knowledge or a person that you wish would write the next book and what would be about.

CLAIRE HUGHES JOHNSON

You know, I don't have like something off the top of my head. I actually think what Satya Nadella has done at Microsoft is incredibly fascinating and just inspirational. I would love to hear the inside story of that. That would be the first thing that comes to my mind. The other would be for those who know me, a pretty classic Claire answer, which would be there's a lot of books by sports, like coaches that, again, have a lot of platitudes and stuff. They aren't as interesting as what I think actually happens in the best teams. And I would love someone to do and you could argue that Michael Lewis has done a little bit of this with like moneyball and what have you, but I really want to get in that. Bill Belichick, my friend Sid, who I had coffee with today, planted this idea with me. Take Bill Belichick's head and then the level of detail in my book and put them on paper. This would just be interesting to me. How do you repeatedly win with really different okay, you have Tom Brady but a lot of other different players over many decades that's anyway, another one. Hi, I don't know how you're going to get that mic.

ELAD GIL

We can repeat the question.

CLAIRE HUGHES JOHNSON

Can you talk a little bit about your experience both in terms of the joys and some of the pitfalls that we should yeah. International scale is a favorite topic of mine. I worked on it at Google and at Stripe. And I think that the really understanding so Google, the AdWords product is actually fundamentally the same and the buyer and what they're interested in, marketing, all of that is actually quite similar in different parts of the world. And I did travel the world and talk to all of the buyers and I was like, oh, okay. We can actually scale a lot of stuff quite similarly in these markets. Of course, you need to localize in terms of language and in some of the ecosystem around it. Right. Some countries have a lot of agencies, some don't. Whatever payments is actually not like that. Right. So you have to understand there's a different regulatory environment, there's different partners, there's often different cultural norms around money and how exchanges of value. And so you really have to think about, okay, what's my balance between central, like federal versus state or central versus distributed understanding of the market and decision making and strategy for a market. And Stripe was much more strategically interesting and complex, and still is, than Google was in that regard. And so then you have to think, well, how do I go back to that delegated decision making in terms of what needs to be empowered locally? How close do the product and technical teams need to be to the markets? And we've tried various experiments at Stripe to get them closer. It's hard though, because a lot of our core stuff is still built like the core payment stack in the US. Right? And it's not easy. And so we're looking at architectural changes we need to make to enable that, because you have to think at all levels, your strategy, your go to market, and your core architecture and product work, how do you distribute it appropriately at the level of where the decision making needs to happen? That's the easiest thing to say and the hardest thing to do. I think the other is this is happening for every distributed company, but how do you enable people to work asynchronously so that they're not dependent on people in other time zones to really get their work done? And does that mean you have to locate more functions? Does that mean you have to just do a better job of documenting, this is how you do the thing, or this is how we handle it in off hours? I don't know. I think you need to be more intentional about your sort of work day to day and how to enable people so they're not waiting for someone else to do it. And then I guess, finally, there's just nothing that replaces going in the market and feeling it and talking to customers. And I think the pandemic really did a disservice because so much international travel is still kind of disrupted and feels like egregious. And I think the best sort of year I had at Google professionally, I mean, personally, very hard for me, but I traveled like almost at least one week a month. But I think I had teams in twelve different offices around the world and I learned more in that year than I've learned. And I just would encourage that for people because you just can't feel it till you're there. But it's a challenge. Oh, the final thing is it goes back to the carving out. If you're having a small country compete for a resource against a large country and the core product, it never wins. And so your strategy, you say all you want about being international and expanding internationally, but unless you have a way to not create that dynamic of the small market loses, you're not actually a global company. And I don't think many companies have solved that because your prioritization process just doesn't work.

AUDIENCE QUESTION

Cool. Maybe up in the back, in the corner. Hi. My name is Gerkarin. Thank you for all the insights and a great talk. I actually work at Microsoft and I've watched the culture transformation firsthand over the last few years. So now I'm feeling inspired to write a book or at least a blog.

CLAIRE HUGHES JOHNSON

Definitely get in there.

AUDIENCE QUESTION

My question was about your career advice. You mentioned a lot of people pay attention to what's hot and it's fairly clear what that is right now. But your suggestion was skate, where the puck is going, what's going to be hot. I'm curious what your perspective is on what that's going to be or what companies that's going to be.

CLAIRE HUGHES JOHNSON

Oh, the prediction. I feel like a lot this is more your category than we're just going to say Generative AI and then name a few companies. I don't know.

ELAD GIL

I think actually sometimes the thing that's hot is correct. And so I think where people get it wrong is when they think the contrarian answer is default correct. And sometimes the mainstream answer is the right one. And if you just ask there's often this question you ask of why now? Like what's shifted? And there's so much that's shifted from a technology perspective that there's clear things to do. And so I do think that that is the biggest opportunity right now in terms of just like right.

CLAIRE HUGHES JOHNSON

But I do think what's interesting about Generative AI is that incumbents, if they actually jumped on it, have their own corpuses right, of data and content. And if they came to the table and then combined that with some of the models that are openly available, you could do really interesting things in a large established company in a way that I think you'd get out disrupted in a different era. And in this one you could in fact win. But you have to shelve a lot of your old thinking, which I'm worried they're not going to do.

ELAD GIL

Yeah. And basically I think it's somewhere between what happened in the Internet and what happened in mobile and internet, 80% of the value went to startups and you had Amazon destroy bookstores and you had DVDs go away and you had all these things shift and Travel equity media changed completely. All this stuff changed for mobile. There was a handful of companies that were started that were new and took advantage of mobile affordances, Uber and Instacart. And Instagram. But most of the incumbents just transitioned to mobile.

CLAIRE HUGHES JOHNSON

Right?

ELAD GIL

They just added a mobile app and they were fine. And I think this is somewhere in between and I think there's a lot of trends that are fake trends.

CLAIRE HUGHES JOHNSON

Right.

ELAD GIL

And so that's the hard part. Ed tech in some sense was a fake trend in terms of actual impact and revenue and things like that, and certain aspects of digital health was a fake trend. And so you sometimes have to ask, will the market structure prevent things from actually happening too?

CLAIRE HUGHES JOHNSON

Yeah.

ELAD GIL

Cool. Maybe right there on the corner.

CLAIRE HUGHES JOHNSON

Hi. Thank you. Thank you. We straddle the intersection of tech and life sciences, and it's been really interesting. To see just the sort of impedance. Mismatch between the different cultures. I was curious to ask, like doing this book tour, you're going and meeting a lot of different people and wondering about whether you think having done that, whether you're thinking the frameworks in the book are universal or whether there's some dimensions that you can have. Like a really high functioning company that just operates in a different place in terms of operating patterns.

Yeah. Now that I'm back in the boston area, biotech and life science is incredibly dominant industry there. I'm spending a lot of time out of my depth with some of these, and the answer is I actually think a lot of it is translatable, but there are more embedded and ingrained. The great thing about tech, we're like a young industry. It's not really an industry. It's like, kind of weird, but so many companies are rapidly evolving and adapting and learning and open to that and young enough that they're not stuck, whereas biotech and life sciences is kind of interesting, right, because there's the large pharma player. You get an industry that has a lot of long term, long tenured DNA. I mean, the laboratory development actually, I think mRNA is going to be such a disruption, right, because it's like, no, actually, algorithms can develop drugs not the way we used to, but I would argue that the longer or the deeper the historical thread of how we do things, then the harder it is for them to see. Oh, actually, you could use a lot of content about what I would think of as more modern leadership or management practices, but they're just having harder time getting there. But they kind of get it. I mean, I went and talked at some MetLife event with these 400 VPs, and they were like, yes, we do need to do things differently. Like, holy crap, what the people we're hiring right now expect from us is not what it was 30 years ago. And I mean, I'd like to I had someone say to me, like, high output management is great, but it is a little bit dated, right. Elements of it are quite sort of manager versus worker and kind of hierarchical or patriarchal even. And that's not my vibe, and so I think they're applicable, but I don't think everyone's open to it is the bottom line. And I also think that in the same way that engineers can get treated as a lot of special snowflakes, congratulations, all of you. The same with the life science researchers, right? And they get away with murder, have your labs. Congratulations. But I think they should maybe learn some new ways of operating and collaborating. $0.01 as an opinion right there.

AUDIENCE QUESTION

So you talked about your early time at google and how the culture there back in the day was very different from what it is now. Can you talk about how the pendulum can shift in terms of the culture of a company over its lifetime and what are some factors that might influence that.

CLAIRE HUGHES JOHNSON

Yeah, I mean, I think the main thing on the pendulum shifting at Google is I'm interested in your opinion was I felt like there was a moment where we were actually scaling the culture pretty well up to a pretty large size, like past what I would have expected. I'm like 15,000, 20,000 people. I think something happened where the leadership team had a meeting or an off site I'm serious. Where they realized, holy crap. This is going to be 100,000 person company easily. And we are not growing fast enough for our opportunity. And we're going to have to bring in a bunch of new leaders, and they brought in a ton of experience across a ton of different functions without really onboarding them into the culture or expecting they should. And it just started to pull into a more tribal, leader based political environment. And I think some things have changed since then. By the way, neither of us have been there for a long time, so we can't really represent Google today accurately. And I know very many great people who are still there, but I think there's been some swings that were about eras of various leaders being brought in, not all of whom were sort of culturally. And I hate using the word culture because you can't really explain it. It's like when you name it, you kind of kill it. But that's for lack of a better word. But I think the big takeaway for me is onboarding new leaders is more important than the time you spend onboarding new employees, and yet it is the thing that gets the least attention, including especially in big companies, and that's a recipe for disaster.

ELAD GIL

Great. Well, thanks again to Claire for joining us. Thanks, everybody, for coming. And again, everybody should check out the book. I think there's copies at the top. It's called scaling people and it's fantastic. So thanks again.

CLAIRE HUGHES JOHNSON

Thanks, everybody. Thanks for being here. Her thanks a lot.

MY BOOK
You can order the High Growth Handbook here. Or read it online for free.

OTHER POSTS

Firesides & Podcasts

Claire Hughes Johnson: Scaling People
Clem Delangue: Hugging Face, Open Source, AI
Dylan Field: Figma, AI & Design, Education
Reid Hoffman on AI, Big Tech, and Society
Sam Altman, CEO OpenAI
Emad Mostaque, Stability.AI
NoPriors AI Podcast

Markets:

Startup life

A brief guide to startup pivots

4 Ways Startups Fail

Founder Investors and Scout Programs

Co-Founders

Raising Money

Old Crypto Stuff:

AI Safety: Technology vs Species Threats

Elad Gil — Fri, 31 Mar 2023 14:59:09 GMT

There are at least two ways to think about potential threats coming from advanced AI. The conventional view is that AI is just yet another of many tool-based technological advances. Like all technologies, the main threat of AI in this scenario is if a human were to use it for nefarious means, or the AI acts badly and needs to be turned off. The alternate, less discussed view is that eventually AI becomes its own self-replicating sentient digital life form, and then there are concerns of competition directly between AI and humans as species competition (versus mere tool use or abuse). This post explores the two different types of threats and their implications.

TECHNOLOGY LEVEL RISK - AI as tool

The most common popular narrative about the risks of AI is to view it through the lens of a tool or technology that is controlled (and abusable) by people (or the AI itself). This viewpoint posits that many technologies that have been good for humans have also been abusable or risky. We have charged ahead with those technologies (everything from fire to cars to nuclear reactors to biotech) and AI is no different - it is just a tool with risks of abuse but also large benefits that make it worth it.

The argument for AI as tool is probably correct in the short run and subsumed by species level risk in the long run. The Technology Level Risk of AI would substantiate either via a person using an AI to do something nefarious, or the AI acting on its own. It may include things like:

An AI manipulates people during an election causing them to vote for someone based on false information.
The AI causes people to die either via an accident or on purpose based on user directions to do so, for example by shutting down a hospital or misdirecting a train.
The AI hacks and shuts down a power grid.
The AI helps design a gain of function virus that gets released by a lab and causes a global pandemic.[1]
The AI is used for large scale bias, censorship, or the erasure of history to enforce totalitarian regimes or views.
The AI starts to think and act for itself, accumulates money and power via the web, and the starts to manipulate society at scale.

All of these are bad things we want to avoid, and many people can be hurt. Some actual large scale disasters could occur. However, these are all survivable scenarios at the species level. In these scenarios, AI stays as in a digital tool form and the worst case scenario is we go in and shut down all the data centers and all the computers in the world. This may put humanity back but we can adapt and restart in the worst case.

The people who make arguments that this is the only risk of AI tend to say things like:

“It is just matrix math. Why are you scared of math?”[2]

“Its just a statistical model / next word predictor!”[2]

“You can always unplug the data center.”[3]

“Every tech wave has risks and we have been fine so far”[3]

All of these are correct statements in the short term. Some of these views tend to trivialize or push aside longer term risks (see below).

SPECIES LEVEL RISK - AI as digital life with physical form

For AI to move from merely another technology risk (in the long line of tech risks we have survived and benefited from on net) to a potentially existential species-level risk (all humans can die from this), up to two technological breakthroughs need to happen (and (2) below - robotics, may be sufficient):

1. The AI needs to start coding itself and evolve: tool → digital life transition. There are multiple companies now focused on teaching an AI to write its own code, with the idea that the AI can make smarter and smarter versions of itself, working on its own. This is not a theoretical - in files where it’s enabled, nearly 40% of code is being written by GitHub Copilot, and multiple teams are actively working on an AI that will write the next more advanced version of itself.

Imagine if this applied to you as a person. You could make as many copies of yourself, but each copy would be smarter, faster, stronger than the prior one. These copies could then also make better versions of themselves etc. In a short period of time you would be the smartest, most powerful person on the planet and could take over financially and other ways.

Once an AI can make smarter and smarter, self replicating copies of itself, that can do the work of software engineers or other people, you have a form of digital life. The AI has made the leap from a technology or tool to an evolving digital organism. These organism will be under strong selective pressure - either to do a task or to aggregate resources.

Once we have AIs that code themselves and spawn and evolve on their own, natural selection can take over. Many forms of AI in this scenario should be benign and non threatening. Others should evolve to seek ever more replication, data center utilization, and energy utilization. These “greedy” organisms should be reproductively more successful than the more benign forms, just as has happened as life evolved for the first time on earth. In general, things that reproduce well and grab resources tend to do well evolutionarily and re-enforce that behavior (irrespective of their task at hand)[4].

Anyone who has worked in biology, particularly on evolution, knows that some evolutionary epochs can happen very quickly. The hard part in biology is that atoms tend to be slow and reproduction may take some time. The great thing regarding digital systems is how fast you can iterate and build things. This should apply here as well. Once liftoff happens in self-coding self-replicating sentience, it should change and evolve and grow a digital ecosystem on a fast exponent.

Once AI become a self-evolving life form, the ability to control what, or how, it evolves against utility or selection functions become hard or impossible. Just ask any biology grad student who has evolved a protein or other molecule and seen the weird, emergent artifacts that come of such experiments. Evolution, unlike engineering, is quite messy.

In this scenario, the AI has yet to control the external physical world and largely manipulates atoms via humans. If all (or most) humans die, the AI is screwed as well.

2. Robotics need to advance: Digital→real world of atoms transition. Currently robotics are in a reasonably primitive state, with much of the emphasis on things like automotive assembly or liquid handling for biotech, as well as drones are used for simple defense and other applications. At some point we will use advanced AI systems to help us design robots/drones to do dramatically more impressive things. There will be strong economic drivers to adopt robots for many tasks - for example robots to build our houses, our data centers, our solar farms and power plants. Robots and drones will also be used (and are already used) for things like security, defense and remote monitoring of pipelines or infrastructure. Advanced robots and drones will probably be best controlled by advanced AIs that have been designed or evolved/grown to control these robotic bodies to do work on humanities behalf. We are also increasingly invest in things like drone based defense, ships, and cars[5]. Eventually, robots will be advanced enough to build their own factories, from which more machines may be produced.

The building of advanced robotics controlled by AI may be the place from which it is hard to turn things back. Once AIs have a physical form and the ability to fight over and control the physical world, it will be much harder to “simply turn off the data center”. Eventually there will be the need to navigate tradeoffs in resource allocation between species. Should the lithium mine be used by humans or AIs or shared? Should land be used to build cities and farms for people to live and have children or should the surface of the earth be covered by data centers and power plants so the AI can keep growing and replicating? This is the type of question two species would need to negotiate and navigate, versus a tool and its owner simply dictating a solution.

One can imagine early on resources will be less constrained and trade and cooperation between specific can thrive. Eventually resources or incentives become misaligned, and competition and conflict breaks out. The main question is what is the timeline to this key set of moments in time?

Approaches to counter species-level risk

There are a few approaches to counter species-level risk, and may of these approaches may work in the short run but not long run, or vice versa. Some may not work at all or may be wishful thinking.

Registry and regulation. Some advocate for registering AIs above a certain complexity or capability set, or to openly register experimentation in them. Precedents include everything from registering and getting approval on the use of nuclear material, on through to biotech biosafety level ratings, where only specific labs with specific ratings are allowed to do certain types of research or drug development.

Alignment research. The minor forms of alignment collapse into politics and censorship. The major forms of alignment focus on human/machine alignment and ethics. Just as someone raised Catholic may have a different moral worldview and framework from someone raised in an anarchist collective, the way AIs are trained and their learning reinforced may impact how they interact with humanity. An example may be Anthropic’s work on Constitutional AI. Once AI starts evolving a reproducing initial conditions may matter less than utility functions and Darwinian selection. So alignment may be beneficial for technology risk more so than species risk (although both may benefit).

Ban advanced robotics. If one believes the premise that the AI rubicon is when AI can manipulate the physical world via advanced robotics, then banning advanced robotics would be one option to avoid species level risk. Of course, there are strong economic headwinds to this idea - for example robotic surgeons to help patients, robotic construction of houses to make them more affordable etc.

Merger. A number of people in the AI community argue the approach to prevent AI from taking over is for humans to merge with AI via brain-machine interfaces or human upload. It is unclear why an AI would actually merge with a human, so some argue we should “force them to merge with use while we can still force them to do so”. Merged people would then effectively have extreme intelligence and digital super powers. The AI researchers who advocate this often think of themselves as the first people to undergo these mergers. If one can become a diety, why not?

Focus on incentives. Depends on the incentives the AI has and the utility function it evolves against, there may be mutually beneficial outcomes. Would AI do better on another planet? Specific geos? Other? How can incentives be aligned?

Hope for the best. Sometimes the future is hard to predict. The old saying in tech is that less happens in 2 years than you expect, and more happens in 5 years than you expect. Maybe you can just relax and enjoy life.

Other. There are undoubtedly many other approaches smart people have considered.

AI will do a lot of good, before the existential threat emerges as a possibility

The hope of everyone working on AI is that AI will be a huge net positive for humanity. Humans around the world should have access to the most advanced medical knowledge and software-driven doctors, should have a universal translator to allow us to communicate with anyone anywhere, AI to advance science and help us make basic discoveries in biology, math, and physics, and should help us understand the world and navigate our relationships. Just as previously technology shifts also led to mass human displacement, that is likely to happen here. We have survived those shifts and overall benefited from them as a global species.

If AI was merely building tech tooling, there would not be a basis for existential concern. However, once AI is a digital self-replicating, self-evolving system with its own evolutionary drivers and utility functions, and external substantiation in robotics, we will have a new sentient species to deal with. The hope is we will have a collaborative relationship, but we can not simply assume that to be the only, or the most likely, potential outcome.

Why write this post?

One of the reasons I wrote this is to capture a framework (and force some thinking for myself) that has helped me differentiate between risks from AI. Too many people talk past each other on this issue as they mix “yet another technology that has some risks” with “AI will evolve into an organism and kill us all”. Many people have different underlying assumptions on what they are actually talking about, and often talking past each other.

It also is important to consider what are the one-way doors humanity can walk through from which it can not return? Creating physical forms for AIs strikes me as a one-way door, and the last step which is incredibly hard to back off from once it happens. It is useful to think about these areas now as we create new sentient life for the first time. AI is already smarter than people on many dimensions, and is likely to eventually be smarter at roughly everything. There is a lot of good to come from this AI wave across medicine, commerce and other areas. Hopefully we can navigate this in a way that leaves our species intact many decades from now.

Notes

[1] They can call the AI that designs the virus fAucI.

[2] Sorry to break it to you, but your brain is just running a mathematical model too.

[3] True statements.

[4] This is one part of the paper-clip scenario, which has spawned a fun game. The fancy way of saying it is Instrumental Convergence.

[5] Self-driving cars have been notoriously slow to adopt, in part due to safety concerns

OTHER POSTS

Firesides & Podcasts

Markets:

Startup life

A brief guide to startup pivots

4 Ways Startups Fail

Founder Investors and Scout Programs

Co-Founders

Raising Money

Old Crypto Stuff:

Video and transcript: Fireside chat with Clem Delangue, CEO of Hugging Face

Elad Gil — Wed, 29 Mar 2023 21:21:03 GMT

Transcript of talk below video

ELAD GIL

Hello. Thanks everybody for coming out tonight. It's a pack house so I think we had something like 1000 people who wanted to attend and so I think people are both very excited to see Clem and then I think there's ever growing enthusiasm for AI. So thanks so much for making it. And I'd also like to quickly thank Edwin Lee, ali Pavillon, emily The Stripe AV event Security food and Catering team. Thank you so much for putting on this event tonight and hosting everybody. We're going to be talking about Clem's background and origins and so I'll keep the intro really brief, which is Clem is the CEO and co founder of Hugging Face, which is really one of the main pieces of infrastructure that everybody uses in the AI industry. He's been working on AI for about 15 years now, originally from France, has been in the US for about ten years. And so welcome and thank you so much for joining us today.

CLEM DELANGUE

Thanks for having me. Excited to be able to chat.

ELAD GIL

Okay, and so could you just tell us a little about the origins of Hugging Face and how you started working on it, what it was originally, how it morphed into what it is today, and how you got started.

CLEM DELANGUE

Yeah, absolutely. As you said, I've been working on AI for quite a while. Before it was as sexy, as hot, as popular, as mainstream as today. And I think that's what gathered our cofounders with three co founders for hugging face around this idea that it's becoming kind of like a new paradigm to build technology. And we were really excited about it. When we started the company, we wanted to work on something that was scientifically challenging, because that's the background of one of our co founders, Thomas, but at the same time, something fun. And so we actually started by building an AI tamagotchi, something like a chat GPT, but really focused on fun and entertainment. At the time, there was Siri Alexa, but we thought it was pretty boring to focus only on productivity answers, and we actually did that for almost three years. We raised the first precede seed on this idea. Some users really liked it. Actually. They exchanged a couple of billion messages with it, but kind of like organically, and I can tell the story later. We pivoted from that to what we are right now, which is the most used open platform for AI.

ELAD GIL

What got you interested in AI to begin with? I mean, you started 15 years ago working in the area, and I feel like AI has gone through different waves of popularity.

CLEM DELANGUE

Right.

ELAD GIL

We had Alex Net sparked a lot of interest there's a CNN and Rnn world. Did you start even before that, or when did you first get interested?

CLEM DELANGUE

Yes, at the time, we weren't even calling it AI or machine learning. The first startup I worked for was a company called Mood Stocks, and we were doing machine learning for computer vision on device. So we were building a technology to help you point your phone at an object and recognize it. And even at the time, it was kind of like mind blowing what you were able to do with it. I remember, I think for me, the realization of how AI could really unlock new capabilities is when I met the founders of this startup. I was working at Ebay at the time, and they told me, oh, you acquired this company called Red Laser that is doing barcode recognition for you to recognize objects and then kind of like, pull up the Ebay page. They told me, you guys suck. You should use machine learning. And instead of recognizing the barcode, you can actually recognize the object itself. I was like, you're crazy. It's impossible. You can't do that with kind of like traditional software. You can't do that with code or too many objects. Possibilities are just too broad to do that. And they were actually managing to do that with some form of machine learning at the time. So that's when I realized, wow, you can do so many new things with this new technology. Apple really led me to where I am today.

ELAD GIL

That's cool. So you then started hugging face. You're going to do like an AI tamagotchi. And I think it's funny how you used to say AI and people would sneer at you and they'd be like, no, it's machine learning. Right? And so I feel like the lingo has shifted back to AI again, given what some of these systems can do, and then what made you decide to move in a very different direction of what hugging faces?

CLEM DELANGUE

Yes. It was very organic with one of these founding moments. It's a good thing that we had stripe because I think it's Pat Kodisan who talked first about the importance of not just founding a company, but having founding moments that change the trajectory of your company. And for us, that happened thanks to Thomas Wolf, one of our co founders, who I think it was like a Friday, Friday night. It was like, I've seen this thing called Birds that was released by Google, but it kind of sucks because it's on TensorFlow. I think I'm going to spend the weekend porting that into Pytorch. And we're like, yeah, you do. You have fun. Have fun during your weekends. And on Monday he came back and it's like, okay, I'm going to release it. And he released it on GitHub, tweeted about it. And we got like 1000 likes, which for us at the time, we were like, Nobody's French. Nobodies. We're like, what's happening? Why are people liking this very specific, very niche, very kind of technical tweet about Pytorch port of bird. There's something there. So we kept kind of, like, exploring that. We joined them, started to add other models to the GitHub repository. And the community came together. People started to fix bugs for us in the repository. We're like, Why are people doing that? They started adding models. Right? They started, for example, the first GPT. They added the next models that were released, and really fast. We ended up with one of the most popular GitHub repository for AI. And that's kind of like what transitioned us from this first idea to where we are now.

ELAD GIL

Okay, and could you describe for people who I'm sure most people know, but could you describe for people what hugging faced us today and how it's used and the importance of the product and the platform and the ecosystem?

CLEM DELANGUE

Yes. Now we lucky to be the most used open platform for AI. You can think of it, as mentioned before, some sort of a GitHub for AI. So the same way GitHub is this platform where companies host code, collaborate on code, share code, test code. We're the same way, but for machine learning artifacts. So there's been more than a million repositories that have been hosted on the hugging face platform with models, most of them open source. So maybe you've heard of stable diffusion. T five Burt originally, obviously, Bloom, for example, Whisper for audio data set. There's over 20,000 open data sets that you can use on the platform. And demos. Over 100,000 demos are hosted on the platform. And more than 15,000 companies are using the platform to bring AI into their features, into their products, or into their workflows.

ELAD GIL

Yeah. Some of the most popular questions on Dory or through the Airtable forum that people asked were around the future directions, because given the centrality of where hugging face is, there's so many directions that could go in everything from like, bespoke B, two B hosting, to tooling, to other types of products or activities. What are some of the major directions that you folks are pursuing currently in terms of product?

CLEM DELANGUE

I would say there are two main directions that we're following right now. One is, like, we're seeing that AI is turning from kind of like these niche techniques, solving some problems, to the default paradigm to build all tech. And for us, that means going from text that is really used on the platform right now, audio that is also really used, and text to image, to expand to every single domain, right? So, for example, last week we've started to see the first open source text to video models, right? We started starting to see on the platform a lot of time series models, right? Like, to do financial prediction, to do, like, your ETA when you order your urban. We're also starting to see more and more biology chemistry models. So kind of like making sure that we support these broadening use cases for AI is one, and the second one is making it easier for everyone to build AI, including software engineers. Historically, our platform has been more like designed for machine learning engineers and people who are really kind of like training models, who are optimizing models, assessing models. What we're seeing now, especially with the AI APIs, is that everyone wants to do AI, right? Even complex software engineers, product managers, infrastructure engineers. So a big focus of ours and some of the things that we've released in the past few weeks and that we'll keep releasing is kind of like reducing the barrier to entry to using our platform. Because ultimately, we think every single company or every single team should be able to use open source to train their own models, right? Everyone is talking today about chat GPT. About GPT. Four. But I think in a few months or in a few years, every single company is going to build their own GPT Four and they're going to train their own GPT Four the same way today. If you think of it, every company has their own code repository, right? And there's as many code repositories as companies. We think tomorrow every single company is going to have their own models, their own machine learning capabilities, not really outsource it to someone else, but really have these capabilities that will allow them to differentiate themselves, to cater to their specific audience or their specific use cases.

ELAD GIL

It's interesting because when you talk about the future, one thing that I'm really stricken by is if I look back over the course of my career, there have been multiple or a small number of very large paradigm shifts or platform shifts. Right. So there was the Internet, which was obviously a huge transition in terms of bringing everybody online. Then a few years later, we ended up with mobile and cloud. So suddenly you could host anything anywhere and simultaneously people could access any product from anywhere in the world. Crypto, I feel, was almost like a side branch that went down the financial services route but didn't become a true platform, at least not yet in terms of compute. And then now we have AI. And it feels like with each platform shift you have three or four things that change. Right. The input and output of how you program a system shifts in some ways, or at least the types of data you deal with, user accessibility and UI shifts. How do you actually interface with something from mobile was different from the desktop? And then the size and magnitude of the implications of that shift are massive. Right, and so if we view AI as a new platform, how do you view or how do you you mentioned everybody will have their own form of GPT four. It seems like the nature of programming itself may change at some point and we can put aside the whole question around do we also create a digital species? And maybe we talk about that at the end. But how does hugging face aim to play a role in terms of this massive transition of platforms?

CLEM DELANGUE

Yeah, the way we see things is we really like Andreish Kappati analogy of Software 1.0, which is the way and the methodology that we've been building technology with for the past 15 years. And now AI is Software 2.0. Right? It's a new methodology, it's a new way of building all technology. It's a new paradigm, the new default to build all technology. And if you think of that, you need for this new paradigm better tools, more adapted tools to do that. And you need better communities, you need ways for teams to collaborate and for the whole ecosystem to collaborate. And that's what we're kind of like trying to provide like a new tooling, a new collaborative platform to build AI better. And we're also trying to build a future that we're excited about. I think a lot of people are kind of scared about AI right now and the potential and the risks associated to it and the way we think about things. If you can build a future where everyone is able to understand AI and build AI, you remove a lot of these risks because you involve more people. So you reduce, for example, the probability of very biased systems. You give the tools for regulators to actually put in place safeguards and you give companies capabilities to align the systems that they use and provide to their users and their customers with their values. Right. Which is what you want. Ultimately you want Stripe to be able to say, this is our values, so this is how we're building AI in alignment with these values. So that's also something important that we're trying to do. We say sometimes that our mission is to democratize good machine learning and we're working really hard on that because we think it's important for the world.

ELAD GIL

Yeah, it feels like Hugging Face has always been very consistent in terms of wanting to have ethical AI or ways to participate that are strong in alignment. A number of companies, like for example, Anthropic, has this approach of constitutional AI where they basically say, we almost provide a constitution as we train the model for what should govern the activities or actions of the model that results. What are the approaches that you think work best and what do you hope that people are doing more of relative to alignment?

CLEM DELANGUE

Alignment is this kind of like complicated terms because it means different things to different people. It can be taken from the ethical standpoints in terms of alignment between values and systems. A lot of people use it today as more kind of like accuracy improvement. To be honest, when they kind of do some alignment work, they actually make the models more accurate thanks to reinforcement learning with human feedback. So it's kind of like hard to debate around that. I think in general, in my opinion, you can't control, improve and align a system that you don't understand. So the main thing that we're trying to push at Hugging Face is more transparency in terms of how these systems are built, what data they are trained on, what are the limitations, what are the biases. And I think if you create more transparency around that, you can almost create a system that is more ethical at core. So that's kind of like the biggest thing that we're focusing on.

ELAD GIL

What is your biggest concern in terms of how open source AI could be misused or abused?

CLEM DELANGUE

There's a lot of things that can be dangerous with AI, however it's distributed through APIs or open source. The biggest thing is dual use when you want to kind of use it in a way that is not the right way, that Model Builders defines. And so one thing that we've been experimenting with, which is super early and probably not a solution to everything, is creating new forms of licenses for models. So we've been supporting something called Rail and Open Rail, which is responsible AI license, which is supposed to be an open license for everyone to be able to use the model, but that defines uses that are prevented from the Model Authors as a way to create kind of like legal challenges for people to use it the wrong way. That's kind of like one approach that we've taken to try to mitigate some of the dual use of AI in general.

ELAD GIL

I guess as you look at the world of open source versus closed source, one of the things that's really been happening is when, before many of the different industrial research labs, the Googles and the opening eyes of the world would publish a model, they'd actually also publish the architecture of the model. They'd publish a paper that goes in depth in terms of how the thing works. The original transformer paper was reasonably explicit, and now they're starting to curtail the amount of information that's coming out with each incremental model. Do you think that puts open source at a disadvantage or how do you think about the future, particularly on the large language model side? Because when I look at the image gen models, they tend to be reasonably inexpensive to train. They tend to be more open source heavy. And it really seems to be more along the lines of the foundation models where this could become an issue because of the ones that need massive scalability and compute. Are you concerned about the lack of publishing? That's starting to happen. And how do you think about the delta between open and closed source models for big foundation models?

CLEM DELANGUE

Yeah, it's definitely a challenge. I think it's good to remember that we got where we are today thanks to open science and open source. Every system that is around today is built stands on the shoulders of giants. Right? If there wasn't research papers for Birds, for Transformers, for 45, for GPT, maybe we would be like 50 years away from where we are today. I think that's what created this massive positive loop that made the progress of AI, I think, faster than anything you've seen before. And if we stop doing that, it's going to slow down. It's going to take more time, and we'll just kind of move slower as a field. But I think one thing that we're seeing is that life abhors vacuum. I think that's the proverb, right? So I think if some companies and some organizations just start to do less open research or less open source, what we're seeing is that other organizations will take over and actually reap the benefit of it. So, for example, we're seeing a lot of collective decentralized. Collective there's like a Luther AI that announced the nonprofit a few weeks ago. You have organizations like Allen AI in Seattle. You have organizations like Stability AI runway. ML you have academia that is coming back in the picture. Right. The original Stable Diffusion was built in German University in a group, a research group called Comvis using Stanford, doing more and more in open source and for open research. So I think ultimately that's what we're going to see. We're going to see different sets of organizations taking over and kind of like contributing to open research and open source because at the end of the day, it's not going to go anywhere, right? I mean, if you look at traditional software, there's always open source and closed source, right? And open science is not going to go anywhere because the goal of most scientists is actually to contribute to the society and not just to do something to make the company money. So I think that's what's going to happen. Maybe like the types of companies that are doing open research and open source are going to evolve, but I'm not too scared about it. One kind of like proof of that is that the number of models and open source models, number of open source data sets, number of like open demos on hugging face has been actually accelerating for the past few months. And you're right in pointing out that we're a little bit biased on text, right? That's one area where proprietary is ahead of open source, right? Large language models. But if you like look at audio, kind of like the best things are like Whisper for example, thanks to OpenAI that is open source. If you look at text to image, stable diffusion is huge and probably like bigger than any puppetry system. If you look at biology, chemistry, time series, also open source is very powerful. So I think it's always some sort of a cycle, right? Sometimes puppetry gets ahead thanks to some companies that are doing a really good job. Like OpenAI is doing an amazing job, for example, right now. But sometimes open source catches up, sometimes it's going to be ahead, sometimes it's going to be a little bit later. That's kind of like a normal technology cycle, I would say.

ELAD GIL

Yeah, I think that's true. If you look at passive technology cycles, it looks like often the really successful large open source approaches that are offsetting commercial efforts tend to actually have a large commercial backer who wants to offset the activities of others. It's almost like strategic counter positioning. So for example, in the biggest sponsor of Linux was IBM because they were trying to counter Microsoft. And then if you look at a variety of open source mobile browsers, webkit is backed by either Apple or Google depending on the branch. Who do you think or do you think somebody will emerge in terms of becoming one of the major sponsors of open source? Does Amazon do it to offset Google and the relationship between Microsoft and OpenAI in the cloud? Is it nvidia is it Oracle? Is it a conglomeration of multiple parties? Or do you think a government or somebody else may intervene in this case?

CLEM DELANGUE

Yeah, I think there are a lot of big tech companies that have kind of like good alignment with open science and open source. You mentioned some of them. Like Amazon has been really good backer of open source. NZDA has been a very good support. Microsoft has been supporting open source a lot too. So yeah, I think some of it is going to come from there. I'm also excited about more governments involvement in kind of like democratizing access to compute, which has been kind of like one challenge for large language models. So when we trained with the big science group, a model called Bloom, which at the time when we released it was the largest language model that was open source, we got support from a French supercomputer called John Z. So I'm excited to see that more because I think if you look at how public policy and kind of like governments can have a positive impact, I think providing compute to universities or, like, independent organizations, nonprofits in order to avoid concentration of power and create more transparency is a very obvious way where they can have an impact and a positive impact on society. So I'm also excited about that, about this ability for public organizations to support more open source and open research in AI.

ELAD GIL

Yeah, it makes a lot of sense. I guess if you look at the types of open source, there's going to be models of various sizes. And to your point on the large language models, if you assume the rumor of the public estimates are if Gpt-3 took $10 million at the time, although I guess now it would be $7 million to train and then GPT four, say, was 50 to 100 if you were to do it from scratch. And then maybe GPT five is $200 million and GPT Six is half a billion or whatever it is you keep scaling up cost. And so you need these sort of large sponsors to at least be at the cutting edge of all times. But then one model behind may be dramatically cheaper. And so it's interesting to ask how that world evolves relative to government intervention or corporate intervention or other things in terms of sponsoring these models, with the.

CLEM DELANGUE

Caveat that we've seen that some scaling is good. We don't really know if that's the scaling that helps the current emerging behavior, to be honest. And that's one of the challenge of the lack of transparency that's happening right now.

ELAD GIL

Actually a really interesting question. What do you think are the basis for the emergent behavior and what do you think is the biggest driver for scale going forward? Is it compute? Is it data? Is it algorithms? Is it something else?

CLEM DELANGUE

I think we're starting to realize and have a better consensus in the science community that data, and not only the quantity of data, but the quantity of data is starting to matter more than just blindly scaling the compute. But I think also something that gap like is important to remember is that training a very good large model today is still very much an art. And it's not just a simple recipe of saying like, you have good data, you have a lot of compute. You're going to get a good model. It's very much still like a very difficult, very hard to understand technical endeavor. It's almost like alchemy that a very small number of people really manage to do today, right? And maybe it's like 20 people in the world today. Maybe it's 50 people in the world today. It's a very small number. I think people sometimes don't realize that. And so I think there's a lot of progress also to be made on understanding the techniques to get to a good model almost independently of compute and data.

ELAD GIL

Why do you think it's such a small number of people?

CLEM DELANGUE

It's a billion dollar question. If if it was easy to to know, I think everyone would would be doing it. I think I think interestingly. It's a mix of technical skills, science skills, and kind of like almost projects management skills, which are kind of like unique. It's not just a matter of doing the right training, but it's kind of like knowing how much more training you want to do. It's a matter of kind of like understanding when you want to release things, when you want to keep doing kind of like optimizations before launching your training run, when you want to start the big six months, three months training run, or where you should kind of keep experimenting. So it's a mix of all of that, which makes it super hard, but super fun at the same time, right? If it was too easy, wouldn't be fun, but hopefully it gets easier and it gets more democratized so that everyone can kind of take advantage of that, reap the benefits of that, learn from that and then, as we said before, build like better systems for each organization.

ELAD GIL

Where do you think are the most exciting areas of AI research right now? Or where do you wish more people were working?

CLEM DELANGUE

I'm super excited about it's fun to do like text, right? And I'm just here for short periods of time. So I went to a couple of Acathan and there are some really cool stuff. But I think it's interesting and important to work on more technically challenging problems right now, especially in the other domains. Like, I'm super excited about biology. How do you apply AI to biology? How do you apply AI to chemistry? To kind of both have positive impact in the world, but also to differentiate yourself and kind of build a more technically challenging stack for AI. So these are some of the things I'm excited about right now.

ELAD GIL

And then how do you think about I feel like there's two views of the world and maybe neither is fully correct in terms of general purpose models versus niche models. Right? So some people are making the argument which is you just keep scaling up models, you make them more and more general and eventually they can just do anything. And then on the other side of it, people are saying, well, just do the focus small model that is targeted to the specific thing that you're trying. To do with the data set that you're trying to do. It can be highly performant and you don't need to wait for the big generalization. Where do you think we'll be in three or four years?

CLEM DELANGUE

Yeah, that's a good question. I've tried to stop doing predictions in AI because it's too hard these days. Like I say something three months, three months later, it goes completely the other way around and I look like a fool. So I won't do too many predictions, but I usually try to more like look at the past and data points. Since Chat Gpd got released, companies have uploaded to hugging face over 100,000 models, right? And I don't think companies like train models for fun if they can use something else, if they don't need the training, they would. And an interesting, other interesting data point is that if you look at all the models on the hugging face hub, the most used ones are actually models from 500 million to 5 billion parameters. And I think the reason why is that when you get kind of like more customized spatialized models, you get something that is first simpler to understand and iterate on. You get something that is faster most of the time, which sometimes can run on device on your phone or on specific hardware, something that is cheaper to run and actually gets you better accuracy for your specific use case. When you specialize it sometimes for some applications when you're doing a chat bot for customer support, where customers are asking for your last invoice, you probably don't need a chat bot to be able to tell you about the meaning of life and the weather in San Francisco. You just need it to be really good at your specific use case. And what we're saying is that having a more specialized, customized, smaller model for that usually is a better fit. But there are some use cases, like if you're being for example, and you want to do like a general search engine to be able to answer all these questions, obviously, like a large more general model makes sense. Ultimately, I think there's going to always be all sorts of different models the same way there are all sorts of code based, right? Today you don't really say oh, my code base is better than yours. You don't say like Stripe code base is better than Facebook code base, right? They just do different things, right? They answer different problems to different questions. The same for models, there's no one model that is better than others. It's more like what model makes sense for your use case and how can you kind of optimize it for your specific use case?

ELAD GIL

The last set of questions I wanted to ask before we open things up to the audience is around business models and business opportunities. And Ali, the Cofounder and CEO of databricks, has this really good framework for open source where he says with open source. You first start out with some open source software and just making that work is like hitting a grand slam in baseball. And then you put down the baseball bat and you pick up a golf club and you hit a hole in one to have a successful business. So it's almost like you need two miracles in order to build something amazing and open source, sustainable as a company as well as a product. How do you think about Monetization of Hugging Face and what are some of the directions that you all are going for that?

CLEM DELANGUE

Yeah, I don't know if I totally agree with this analogy because I think open source also gives you superpowers and things that you couldn't do without it. I know that for us, like I said, we are the kind of random French founders and if it wasn't for the community, for the contributors, for the people helping us on the open source, people sharing their models, we wouldn't be where we are today. Right? So it also creates new capabilities, not only challenges for us, the way we've approached it is that when you have kind of like an open platform like Hooking Face, the way to monetize is always some sort of freemium model or some kind of like version of a freemium model. So we have 15,000 companies using us right now and we have 3000 companies paying us to use some of our services and usually they pay for additional features like enterprise features, right? Some companies, they want security, they want user management or they pay for Compute, like they want it to run on faster hardware, they want to run the inference on the platform, they want to run the training on the platform and like that. We found kind of like a good balance where if you're a company actually contributing to the community and to the ecosystem, you're releasing your models in open source, it's always going to be free for you. And if you company more like taking advantage of the platform then you contribute in a different way, you contribute financially. Right. By helping us monetize and keep working on this. We're still early on that, but we've kind of found this kind of differentiation between the two that allows us to keep working for the community, keep doing open source, keep contributing in alignment with our values and what we want to do, but at the same time make it like a good business, a sustainable business that allows us to scale and grow our impact.

ELAD GIL

You mentioned the community a few times and I think Hugging Face is one of the most beloved products and communities in the AI world. Were there specific tactics you took to build out that community or things that you felt were especially important in the early days? Or how did you evolve something that's so powerful from a community basis perspective?

CLEM DELANGUE

I would say just the emoji, having the hugging face emoji as a logo as a name. That's all it took to get the love of the community. No, it's hard to say. We're really grateful. Some of the things that we've done that we've been happy with is that we never hired any community manager. It's a bit counterintuitive, but it led to actually every single team members, every single hugging Face team members to actually share this responsibility of contributing to the community, talking to the community, answering to the community. Instead of having a couple of team members and instead of having researchers being like, oh, I'm not going to do the community work because we have this community manager. So, for example, our Twitter account, the hugging Face Twitter account, everyone in the company can tweet from it. So if you're seeing the tweets from the hugging Face Twitter accounts, it's not from me, it's not from like a community manager. It's from any of the hugging Face team members, which was kind of like a bit scary, scary at the beginning, especially as we grow. We haven't had any problem yet. But I apologize in advance if at some point you see like, rogue tweets that might be a team member.

ELAD GIL

But yes, it's a smart approach to always be able to blame someone else for exactly.

CLEM DELANGUE

Maybe it's going to be me, actually who's going to be doing the bad tweets, but I'll be able to say it's an intern or something like that.

ELAD GIL

Yes, you must have a lot of interns just in case you kind of yes, that's smart idea. Last question for me. And then we'll open up to the audience. What do you wish more startup founders were working on? Or where do you think there are interesting opportunities for people to build right now?

CLEM DELANGUE

So I'm a bit biased on that, but I wish more startup founders were actually building AI, not just using AI systems, because I think there's a big difference. The way I see things in the early days of software, you could, you know, use an AP, you could use an API, you could use like a Weeks or Squarespace or WordPress to build a website. Right? And that's that's good. That's kind of like a good way to get something up quickly and you can do beautiful things. But I think the real power came from people actually writing code and building technology themselves. And that's how you get kind of like the power out of this thing. It's kind of like the same for AI, right? You can do something quickly. I think ultimately, if you really want to be serious about AI, you need to kind of understand how models work, how they're trained, how you can optimize them. And that's also what's going to unlock the most potential for truly great startups and great products and companies that are differentiated from incumbents, just like adding some AI features. And we're seeing a lot of these companies like Runway ML announced the release of their text to video, I think today or yesterday. That's a good example of a really kind of like AI native startup that is really actually training models, building models, really kind of like doing and building AI, not just using AI. So that's one thing that I usually recommend startups do. Or if you're just using AI, just build your company accordingly, knowing that your mode or your advantage, especially dirty stage, won't be so much on the technical capabilities, but more on getting customers or getting users or any you wrote a beautiful, very good article that I would recommend everyone to read on modes for AI. They need to then kind of take advantage of other kind of modes than, in my opinion, like technical modes.

ELAD GIL

Okay, great. Let's open it up to the audience. If there are any questions, maybe we start in the corner right there.

AUDIENCE MEMBER

Hi. You mentioned something about open sourcing thing. It's good because you give it to good actor and also bad actor can use. But good actors probably are. We have more good actors. But how do you usually respond to claims like OpenAI that they don't say any details about their models, they don't open source anything because they're afraid of AI safety.

CLEM DELANGUE

I respect everyone's approaches like different organizations have different ways of seeing the future or the current way technology is building. I have a bit of a different view. Right. For me, if you look at development of technology, usually the biggest risks come from concentration of power and the fact that some technologies are built, like behind closed door. And if you build things like in the open, you actually create a much more sustainable path in the long run for this technology to be embedded in society in general, right? For regulators to be able to create the regulatory framework for these technologies, for NGOs, for civil society to be able to weigh in. So, yeah, I think we're starting from a very different position, philosophically speaking. But that's not too much of a problem, in my opinion, for the ecosystem. You can have different organizations with different points of views and kind of like the most important thing is just that your company doing are aligned with your company values.

AUDIENCE MEMBER

I wanted to ask a question on privacy. You mentioned data is integral to improving models. A lot of end users struggle with This paradox of this paradox of how do they preserve the privacy of their data while improving the models? Is in housing open source models the solution or approaches like Federated Learning? Appreciate your thoughts.

CLEM DELANGUE

We've been working a little bit on kind of like more like distributed or decentralized training, which is still hard to do. And I think nobody has really figured it out yet. But that's when you asked me about my interest in the science progress. That's one area where I'm really excited about to see more people working. But yeah, the more like, practical answers and solutions today are on device models, or there are also some more solutions that are embedded in how you train models. So, for example, we're leading an initiative called Big Code, which is releasing some, I think it released a few weeks ago, the biggest open repository of code that people can train code models on. It's called the stack. And the interesting thing about it is that we gave the ability to opt out from these data sets before training the model. I think you've seen last week, the training of the Adobe model that also have been really good at training on good data, where users have actually opted in for the training. So these are also some, I think, important developments in the field where you want to be a bit more intentional about the data or more transparent about it. One of the challenges is that a lot of these systems today, we don't really know what they've been trained on. Right? Because there's no transparency about it. I wish there was, so that we can kind of have a better understanding of what you're capable of doing with which data and then kind of find solutions to make sure it stays like privacy preserving for people. But there's some good development and I think we're making a lot of progress there.

ELAD GIL

Do you think we're going to end up with just like, those robots text right now for search? Do you think we'll have like, AI text or something for sites to be able to opt out of use in AI data sets?

CLEM DELANGUE

Yeah, probably. We'll need to have norms around that, for sure, around consent for AI. I think that's really important. That's really important, for example, for artists, for digital artists or non digital artists, that's important for attribution and distribution of value, right? Because we want people who are contributing to be able to be rewarded for it. An interesting question that I don't think has a good solution right now. But in a world where search is only kind of like a chat interface, what rewards kind of like the underlying creators of the content, right? If I build a website and before I was leaving because I was getting traffic on this website so I can do ads, and if now the results of this website is actually shown on like a chat answer without mentioning me, as a content creator, what's my incentive to create this content? Right? So will people just stop building websites because they basically don't get the attribution or the reward for it? These are very important questions. I think we're just scratching the surface of what needs to be done for things like that to be resolved, but very important questions.

AUDIENCE MEMBER

My name is DJ. I'm the founder of Sync, where AI we use AI compliance warehouses. These foundation models, they're very expensive to trade and that's often out of the region startups. So what are the modes that can be built on top of them? There's data, there's human feedback, there's data, there's human feedback, there's maybe to some degree being good at prompting. And are there any other methods that you feel startups can build? Modes?

CLEM DELANGUE

That's a good question. That depends a lot on your skills, your background as a team, what you're excited about. I think there's a lot to be built around specializing for a domain, a specific domain, a specific use case, specific industry, specific hardware. Right. That's what I'm most excited about, right. Trying to leverage some specific expertise or specific domain specific kind of like problem that others bigger players are not going to be focusing on. As I said, like for example, biology, chemistry, time series, all these domains where you don't see as much activity. I think it's a good way to have, in a way, more time as a startup to build your differentiation and your tech stack to a point where you're not at the mercy of 100 other startups, like releasing exactly the same thing as you did and kind of like losing your edge. So that would be some of my recommendations. But again, I've told about the story of fucking Face, right? How we started with AI Tamaguchi and ended up where we are. So one of the main things is just to start working, start building, listening to what you're seeing as signals. Iterate on things and I'm sure you'll land on something that you're excited about at some point.

ELAD GIL

Okay, great. I think unfortunately we're out of time. I think for the next 55 minutes. Feel free to hang out here. Thanks to Stripe for being such a gracious host. So this is also an opportunity for you to meet other folks who are very excited and working in AI. Thank you again as well. To clem is awesome.

CLEM DELANGUE

Thank you.

MY BOOK
You can order the High Growth Handbook here. Or read it online for free.

OTHER POSTS

Firesides & Podcasts

Clem Delangue: Hugging Face, Open Source, AI
Dylan Field: Figma, AI & Design, Education
Reid Hoffman on AI, Big Tech, and Society
Sam Altman, CEO OpenAI
Emad Mostaque, Stability.AI
NoPriors AI Podcast

Markets:

Startup life

A brief guide to startup pivots

4 Ways Startups Fail

Founder Investors and Scout Programs

Co-Founders

Raising Money

Old Crypto Stuff:

Startup Decoupling & Reckoning

Elad Gil — Tue, 28 Feb 2023 15:20:53 GMT

During the last few years low interest rates and money printing led to a funding bubble in private technology. Many startups received pre-emptive extra “free rounds” and hired teams much larger then their stage or progress merited. In parallel, abundant money and secondaries meant companies that should have shut down or sold kept going (indeed - this was a pre-COVID phenomenon from ~2017 or 2018 on). There was always another round or extension to keep companies without real product-market fit going. Or founders sold secondary stock in multiple rounds instead of selling a company that wasn’t really working.

These trends have resulted in an overhang of companies that either (1) lived without product market fit and survived well past their natural expiration point, or (2) hired way ahead of progress and burned large sums with high valuations and now are stuck with little progress per dollar and a large preference stack.

When will companies run out of cash?

Many companies are likely about to meet a hard reckoning. This is likely to start end of 2023 and accelerate through end of 2024 or so. The likely timing is +/- 6 months. It is based on when companies last fundraised at scale, 2021, and how much runway they raised.

Many companies raised 2-4 years of runway in 2021 (and Q1 22). A company needs to fundraise when it still has 9-12 months of cash left.

For example, below is the cash out dates for companies assuming they raised 2.5 years of cash in 2021. As you can see, in this scenario many companies of this type cash out by Q3 2024.

For 3 years of cash, cash out largely happens by end of 2024.

Adding an extra year of cash (4 years cashed raise) obviously just pushes things out an extra year - so we should expect a tail of companies that raised even more in 2021 to run out of money in 2025. For a CEO not making much progress, it might not be worth waiting that long to make a radical change to one’s business.

There is likely a mix of companies who have raised anywhere from 2-4 years of cash in 2021 that will not grow into prior valuations, and will need to find an exit or shut down. As such Q3 2023 through end of 2024 (and maybe part of 2025) seems roughly when these companies will start to seek exits and or shut down.

There will also be a set of companies that raised a very large amount of money, will cut burn via multiple layoffs, and roughly have infinite runway without much revenue growth. More on these companies below.

The ongoing correction in mid-to-late stage tech is increasingly decoupled from macro

The ride up, and the cracking point, of tech valuation inflation were driven in part by the macro environment of low interest rates and money printing (and probably in part due to the collective mass fever dream of people stuck at home during COVID). Private tech and market macro (inflation, interest rates) are now decoupling in most ways that will matter for the next 18 months for pre-existing mid-to-late stage companies.

Low interest rates plus money printing puffed up public and private market valuations leading to over funding and over hiring.

This also led to inflation.

In response to inflation, interest rates were hiked.

Interest rates determine how deeply you discount future cash flows, particularly for growth stage companies. When interest rates are higher, growth multiples compress and stock prices drop.

The higher interest rates stomped the bubble in public tech company multiples and pushed them back to in range with historical norms.

Multiples in the COVID era were the anomaly, not the 14 years prior. We are not going back to 2020-2021 valuation multiples (or anything close to it) until the next bubble.

Every $1B in market cap = $100M to $150M in revenue growing 30% a year, $5B in market cap = $500M in forward revenue adding ~$150M in revenue a year (!)

Even if interest rates were to drop and SaaS multiples were to rise 50% they would be at roughly 10X forward revenue (up from 6X today). This means every $1B in market cap would need $100M or so of revenue (down from $150M today), growing 30% year over year. A $5B market cap would mean a company with $500M of revenue (down from $750M today) growing 30% year over year. So a company at that scale would need to be adding $150 million *per year* to be worth $5B. A $10B market cap is double that number ($1B in revenue adding $300M a year!). The current period is around historical norms, not an anomaly(!!).

Does a recession matter?

A recession at this point would slow down revenue growth for many companies[1]. This will drop NRR and new revenue and likely increase burn. Some tech startups that are on the margin from a survivability perspective will tip into a slow motion death as their growth slows. A recession would further drop public market expectations and multiples and that will bleed into later stage private rounds. Public market companies are already projecting slowing earnings and growth for 2023.

Irrespective of recession or not, software demand is softening (see chat below). Part of this is driven by overbuild and over purchase of software during the COVID era when both many “digital transformations” were undertaken, but also when capital was cheap and spending was not very ROI driven. With cost of capital going up and much corporate belt tightening, software spend is likely to continue to slow.

No matter, the valuation reset coming for private tech is so stark (50X to 100X ARR in 2021 → 10X) that a recession or drop in spend will only matter on the margin relative to the likely inability to raise money that is coming no matter what. Bloated, low-to-no product market fit companies, and ones that have product-market fit but overhired and overcapitalized will get stuck in different ways.

The macro economy and the startup valuation reset are now decoupling. The reset in private tech will happen roughly no matter what happens in the macro economy.

If your company has good underlying economics, can grow at a good rate, and is not dramatically overvalued, this could be a golden period for you as you soak up amazing talent and land great customers. If not…

There will be 3 types of outcomes -

1. Shut downs

More companies than people currently seem to think, will have no choice but to shut down. While some CEOs mention that “the worst case is a down round”, the reality is investors will look at money burned, progress by the company, and in many cases decide the state of the company and/or preference stack does not merit further investment. There is a large back log of companies that should have failed over the 4-5 years that kept going due to a loose capital environment. Things are going to catch up.

Now is a good time for CEOs to ask both insider investors, and new potential ones, what metrics they would want to see for them to be able to raise (note: these metric may be off in a year as investors are flooded by companies needing to fundraise, or dealing with their existing portfolio companies that will need tons of help).

Model it out - if you can get to profitability or “default alive” or “default fundable” that is great news.

If it looks like there are not many scenarios for a company to make it, now might be a good time to start an M&A process to sell the company, or to do an early shut down.

Your time is precious

Shutting down and giving up is really hard. Founders may have spent many years of their lives on a company that is not going to work. But rather than spend another 2 or 4 years waiting for the bank accounts to drain through multiple layoffs and venture debt draw downs it might be better to accept reality and throw in the towel.

For a founder (and their employees), those 2+ extra years may be amongst the potentially most productive years of their lives. The opportunity cost of not going to work on something better is too high. And to be able to do so without a cap table overhang, with a new team that fits a new product, may be a better way to proceed than continuing to grind on something that is not working or pivoting with the wrong team and cap table.

Investor incentives differ early vs late (and based on their role in VC firm)

Investors may or may not be aligned with a shutdown. Your early stage investors or those with a high performing fund may want you to keep going no matter what. While for them it is a free option, for you it is your life and livelihood! Early stage investors in particular may have used the latest round price of your company to raise a new fund (“look LPs-my investment track record is working!”) and may push back or dread a reset on valuation as it impacts their own ability to raise new venture funds. The personal incentives of the VC on your board within their firm may also impact things. For example, a VC may resist an otherwise rational low M&A exit or necessary down round in order to get promoted within their partnership or prove that things are working for them career-wise.

Other investors may prefer to get back 40 cents on the dollar, free up time, and also potentially free up founder and employee talent to back or work with again.

Think through your investors individual incentives as you consider what to do and hear (and filter) their opinion. A given company in a VC portfolio is part of a portfolio and may function as an option - while for you the company or how you otherwise spend your time is your entire livelihood. And never forget how precious your time is. Life is short, and productive years in which you can take a lot of risk are quite limited[2].

Anticipate 2nd, 3rd, and 4th rounds of layoffs as companies realize their business is not working, fundraising options are limited, and cash is running out. As mentioned above, doing a firesale or shutting down may be a better option in reality (and a harder pill to swallow short term) than a slow prolonged wind down.

Infinite runway companies

Some small subset of companies may also have roughly infinite cash but without a strong business. Maybe they raised $50M, $100M, or even $500M during good times but their business fundamentals are not there. A company that raised $500M and is burning $50M a year may have 10 years of runway but minimal revenue growth. Options in that case are to keep going, buying other companies to try to spark growth, selling to someone else for your cash, an ongoing search for something new (this roughly tends to never work for a late stage company), re-setting valuation (see below), or alternatively to shut down and return cash. A company that raised at 50X ARR will take 9 years growing 20% a year to grow to flat with its current valuation (assuming at 10X exit multiple).

During the 1990s internet bubble a few companies went public and then sat on $1B+ in cash, in some cases taking a decade to wind down. Just because you have infinite runway does not mean you should stick with it infinitely - your life and time are more precious.

2. M&A

Another path is to sell your company. As 2nd and 3rd rounds of layoffs happen there may be fewer and fewer buyers in the market for talent or new assets. After all, if a late stage company just did a big layoff it may be less likely to acquire (non-AI) talent. However, later stage growing companies, or public companies with strategic needs, may be active buyers. The very biggest, healthiest buyers (Google, Meta, etc) will be stymied by a politicized FTC which seems to be blocking M&A for arbitrary political vs rational consumer-protection reasons.

In parallel, companies that are later stage and burning a lot of cash will find it harder to exit to acquirers. Public company buyers are increasingly judged on profitability and capital efficiency. They may be unwilling to absorb the impact of buying a high burn, high growth company in the current environment.

Most large incumbent buyers may buy (at most) one company in a given area in order to enter that market or bundle an adjacent product. If you add up all the incumbents in any given area who could buy a company, there is probably a 10-1 or 20-1 ratio of companies who need to get sold versus landing spots at incumbents. This suggests if you plan to sell your company, you should pursue an exit sooner than later. End of 2023/early 2024 will become a crowded market for companies trying to sell.

It is possible we will see a rise in M&A solely for a company’s cash, which is a thing in biotech, and may also become a (smaller) thing in tech. In this scenario a cash rich company (whose product is not working) gets bought by a cash poor company whose product is working and who just wants the cash. This sort of M&A is effectively raising a round from another company’s founders and backers.

A company with a lot of cash and a high market cap can also try to buy its way into growth by rolling up other startups with better growth potential, or stringing together a set of assets to cross-sell as a bundle. Very few companies will take this path, although many probably should arbitrage a 2021 valuation for strong external assets and/or more cash.

3. Keep going (& in some cases re-setting valuation to realign incentives)

Companies that can keep going will. A subset of these companies will comfortably grow into their valuations and preference stack and do just fine.

Many other companies are at, or above, terminal long term valuations for the company. As an analog Dropbox’s last private round in 2014 was at a $10B valuation. Almost 9 years later the company is worth $7.7B (valuation not adjusted for dilution or inflation). Many companies that raised in the hey day of 2021 are currently worth the most they will ever be worth over all time.

These companies may consider writing down their stock price to (1) provide more upside for future employees (2) reissues stock to keep their teams focused and motivated and (3) make fundraising and going public easier in the future.

A number of companies such as Stripe, Instacart (rumored to have repriced 74% from $39B to $10B), and Klarna (valuation drop of 85% from $45B to $6.7B) have proactively written down their companies’ market caps in order to reset prices for their employees and/or fundraises (either private rounds or IPOs). These companies should be applauded for trying to do the wholistic right thing for their shareholders given that comparable public market companies may be down 70-90%.

For some companies, the outcome for employees (who got issued smaller packages due to high stock prices) investors, and founders may decouple. Founders may view ongoing high valuations as a net positive for the company or themselves, even if it may also negatively impact their employees and investors. This may in some cases backfire as employees leave due to a lack of upside (unless comp reset) or some investors turn acrimonious. “Good outcome” is relative of course - if many companies go out of business in the coming period a stable job and paycheck will go up in value for employees in its own right and a reset is needed less.

Growth rate impacts how long a company may be “stuck” relative to valuation. For example a company that raised at 50X ARR growing 20% will take 9 years to grow into its valuation. As Alex Clayton from Meritech has pointed out, some companies invested at a 100X ARR may in some cases may take 6-7 years to grow into their *last round* price (even with solid growth at 85% persistence a year), to say nothing of being a few 2-3X outcome on that price. The below spreadsheet he ran is how all companies in public markets are eventually valued. A founders big idea and great team, at some point, will largely be converted into numbers around revenue, burn, growth, margin and related by the public markets weighing machine.

Resetting valuation may, in a subset of cases, be a way to realign these things for all shareholders. Everyone forgets that Facebook did a down round, and Square IPO’d under its last private round valuation, only to later grow well past it. As mentioned above, Stripe, Instacart and others have already been reported to do market cap resets to reflect the current environment. Other companies will grow into and past their valuations in private markets and will not need (or want) to reprice.

The downsides of a valuation reset may include additional dilution, some short term employee moral hit at having a lower stock price (this can be offset by new comp and more upside), smaller perceived equity packages to employees, and more dilution with M&A. Plusses may include more perceived (and actual) employee upside on stock, less debate at M&A for “how much is this company worth”, and in some cases (but not all) more alignment between founders, employees and investors.

It is worth noting that roughly every public company (Snowflake, Shopify, Airbnb, and others) just survived a massive “down round” in the public markets - as their stocks dropped 50-90%. The public markets provide ongoing up rounds and down rounds on a daily basis, and yet these companies work just fine. Some companies may also eventually IPO under their last round prices, resetting the cap table.

Zero sum thinking

Tough times will in part become a test of character. As company after company goes through layoffs, failed fundraises, or outright fails we will see many different types of behavior both good and bad, from investors, founders and employees. It is good to remember in periods of pain that pain is temporary, that technology is a massive wave of change and global economic empowerment, and that many future opportunities are yet to come. It is tough not to get blinded by the moment, but one must strive to see and think clearly and act well in the times ahead.

Notes

[1] Some companies may be less impacted by a recession, such as security products or ones that clearly decrease costs. However, the biggest lever for many companies to cut costs is not software - it might be to lay people off or shut down marginal products. This means even software that increase efficiency gains may see a strong slow down in the short term as their enterprise customers focus on more pressing matters in the short run. Gains may still be had for many enterprises through reasonably brute force actions.

[2] The older you get the more this is driven home. Between kids, sick parents, and other non-work items you realize how free and risk-seeking your younger life can be.

Thanks to Alex Clayton, David George, and Brad Gerstner for comments on this post.

MY BOOK
You can order the High Growth Handbook here. Or read it online for free.

OTHER POSTS

Firesides & Podcasts

Markets:

Startup life

A brief guide to startup pivots

4 Ways Startups Fail

Founder Investors and Scout Programs

Co-Founders

Raising Money

Old Crypto Stuff:

Defensibility & Competition

Elad Gil — Thu, 16 Feb 2023 16:55:04 GMT

There is a lot of discussion of whether building “a mere wrapper on top of GPT” is defensible.

In this post I argue most SaaS software starts off default non-defensible (hence all the HackerNews posts & Reddit threads saying as much), and tend to build a moat over time. In parallel this post delves into different forms of moats as well as competition more generally.

Most early stage startups are not very defensible

Many tech startups will launch an early product within 6-12 months of founding. Teams will initially be 2-5 people with a mix of eng/product/other. Definitionally, it is easy to copy or clone something that has taken a handful of people a handful of months to build.

There are of course counterexamples to this - either in terms of it taking longer to build a technically challenging more defensible product (see eg Snowflake), or there are considerations around IP (for biotech), deals (e.g. payments needing a backend), regulation (“we need to be a broker/dealer first”), talent (“there are 20 people on the planet who can train this sort of foundation model”) or other areas (for example, pre-existing open source software that is being commercialized by the team who created it).

However, the vast majority of SaaS, consumer, and certain types of AI companies are easily and rapidly cloneable. Yet many massive companies have been formed in areas that a priori may be easy to copy or build.

Adding defensibility over time

While a handful of startups start off with defensibility intact, the vast majority do not and need to add defensibility over time. There are a few forms of defensibility that tend to emerge:

Network effect. In network effects every user creates more value for other users, forming a positive feedback loop. Network effects can be local (e.g. the more people who use Slack within a company the more benefit) or global (the more people on Instagram, the more people engage across networks and geos). Few SaaS companies have true network effects, with companies like Github as a potential counter example. Uber had a network effect and positive feedback loop between driver and rider liquidity in a given city. Network effects are one of the few forms of defensibility that can arise immediately upon launch of a company.
Platform effect. Some SaaS or operating system companies benefit from a platform effect. For example, Salesforce is a sticky product because so many other companies have integrated against it. Platforms usually come after a company has been in existence for a few years and has enough users that others want to build against its platform to reach said users.
Integrations. Related to platform effects, sometimes a company becomes defensible via integrations. Integrations may include:
- The company integrates against many other APIs, code bases, etc. that are hard to reproduce (see eg some forms of crypto custody such as Anchorage where each codebase and blockchain needs to be audited for security + added
- Services does integrations for the company against other vendors. For example, ERP systems like SAP and Netsuite are hard to displace as each implementation is a unique, multi-quarter process to integrate the ERP system with many other vendors inside an enterprise. EMRs such as Epic are another example of integration-driven and long-term contract driven defensibility.
Building a ton of stuff.
- Bundling. Bundling and cross selling products prevent other companies from finding a wedge to compete with you, or create better sales success via superior interoperability, pricing, or procurement processes. See Workday, Rippling, Deel or Ashby as examples of companies that early on planned to bundle multiple products usually provided by different vendors.
- Big product footprint. For some products, the breadth and depth of the product footprint means it is hard for new entrants to compete, as it takes too much time to reach feature parity before one can sell to customers competitively. This is why many startups first launch to the low end of a market segment, and then build depth of product over time as they move up into enterprise.
Deals. Different types of deals can create defensive moats. This may include:
- Early access. Early access to APIs, data or other unique assets may provide a competitive advantage. For example, OpenAI has provided some companies with early access to GPT-4.
- Exclusive provider or distribution. Sometimes locking in as an exclusive provider can get scale, brand, or distribution advantages for a company. For example, Google’s early exclusive deals to power search for Yahoo! And AOL were company-making moments. Similarly, Microsoft’s early OS deal with IBM was a king making moment.
- Backend. Some markets require hard to get deals in place or ones that take a year to execute and implement. This may include things like payment bank ends or banking access.
Sales as moat. Many SaaS or enterprise companies eventually develop sales driven moats. Moats may include:
- Multi year contracts. Long term contracts to lock in customers for multiple years. Some industries have extreme versions of this where customers sign up for 5-7 year deals, so once a market is locked up it is hard for a new entrant to gain any share.
- Sales process. Often when selling to a large enterprise you need to go through multiple reviews and functions at the buyer including security review and audit, procurement, IT, and other areas. These reviews can take many months to complete, so it is often easier for enterprises to buy a less-good bundled product from an existing vendor, then a better stand alone product from a new supplier.
Regulatory. Some companies receive regulatory approvals that provide a moat. For example, early on AngelList received a “No Action” letter from the SEC that allowed it to run syndicates.
Data or system of record effect. Having unique or proprietary data, or owning a customers data or having a long historical record of it can create defensibility. Similarly, being a “system of record”, for a user, entity, etc. can be a powerful position to be in. On average, data effects tend to be overstated outside of highly scaled consumer products or very specific verticals.
Scale effects.
- Capital scale. Having access to large sums of money sometimes creates the environment to execute quickly and build a network effect or defensible position - this happened with both Uber and TikTok.
- Business scale and negotiation. Sometimes pre-negotiating “scaled pricing” allows you to offer a service cheaper than anyone else, creating a self-fulfilling loop. This happened with Intel and early semiconductor releases. Similarly, there are scale effects in payments or other services, where the more volume you have, the cheaper your own backend providers become.
Open source.
- Creator advantage. Sometimes the teams that implemented or created an open source software project are best positioned to commercialize said product, given their ability to drive and control contributions to said project + brand and relationships in that community. See e.g. dbt Labs.
- Openness. Some dev teams only want to adopt open-source software for specific areas.
Brand. Eventually a brand may be built that positions a company as the default choice for a specific use case.
IP Moat. Intellectual property tends to be more capable of protecting hard tech or biotech companies than most consumer or SaaS products.
Other forms of defensibility. Usually the forms below create the largest advantage when competing with incumbents versus other startups, as other startups may be able to copy said advantages.
- Speed. One of the few advantage a startup has against an incumbent is its speed. It definitionally will have fewer people, less money, fewer customers and a smaller product footprint. Speed of iteration and execution, responsiveness to customers request, and speed of hiring and closing candidates are all advantages. Oddly, speed also turns out to be an advantage versus most other startups - few teams can execute at high levels consistently.
- Pricing. Occasionally startups may have a pricing advantage relative incumbents (but maybe not other startups) via a lower cost structure, a lack of an existing product to cannibalize, or due to a different business model. In some cases a startup can give something away for free as a wedge to cross sell an incremental product in a way others can not compete. “Your margin is my opportunity” is a famous quote by Jeff Bezos on Amazon and its advantages entering new markets.
- New business models. Sometimes a startup can innovate on business model to create a higher leverage business or different incentive structure. For example, Anduril in defensetech has a traditional tech margin-based business, while all the incumbents sell “cost plus” (ie they charge 5-8% on top of what is cost them to build the product for the Department of Defense). The cost plus model creates a lot of incentives to act badly - for example since labor is charged as part of the cost plus, delays and doing things slowly means more revenue for the incumbents. Similarly, the reason some items have $100 screws is so they can charge 5% on top of it (instead of just using a 10 cent screw).

Most of the forms of defensibility above take a few years to build.

Why aren’t there more fast follows?

Given that most startups take time to build defensibility, this raise the question of why more startup founders don’t just copy companies that are already working, but early in their journey?

Reasons may include:

1. It is sometimes hard to know what is actually working, versus hype.

2. Founders have a lot of pride in what they build, and may not want to just copy and out-execute someone. Often when a startup copies another’s idea, they put a unique spin on that approach or product versus default blankly copying it. All these tweaks and changes tend to make the product worse.

3. Perception that “the market is over” so no one copies a company even if it might be tractable to out execute them.

4. It is harder to hire strong employees to work on what is initially a clone company. People assume more defensibility than tends to exist early, so are harder to convince to join your efforts until traction is clear.

Startups & incumbents - evolution of competition

Most good startup ideas are definitionally non-obvious (otherwise everyone would be doing them). Often, good startup ideas seem small, niche, or toy like. Only desperate people will go and build in these areas as they may seem too small a market or use case initially to large incumbents.

Often in the first 4-5 years of a successful startup’s life, its main competition is other startups. After a company starts to break away from the startup pack, and proves a large market, that is when a big company tends to remobilize and steer into the direction of a startup. This is often done via cross-bundling and cross-selling - for example Microsoft Teams was successful versus Slack as Microsoft bundled and cross sold Teams to its existing enterprise customer base. This allowed it to gain rapid share. Usually this transition from startup competition to incumbent competition takes between 4-7 years. Incumbent market entrants can either be extremely fierce (see Microsoft Teams) or a bit of an afterthought (see Microsoft Bing versus Google, or Google Plus versus Facebook).

In other circumstances, an incumbent may be aware of a large and obvious opportunity and steer into it much faster - we are seeing this happen now in multiple areas of generative AI where incumbents like Microsoft, Github, Notion, and Intercom are amongst the first to act. Similarly, Meta and its sub products Instagram quickly copied SnapChat stories, curtailing its international growth substantially. Fast market entry or expansion by an incumbent can at times be the death knell for a startup, especially if done early (i.e. the first 1-3 years of startup life).

As a proto-founder, it is worth thinking through how aggressive the incumbents are for your potential product area, how clear they are on the opportunity you are going after, and how likely they are to act. You can often find this out simply by meeting current executives at the company, or asking former employees. Earnings call transcripts can also be revealing at times, as can prior bias to action and ability to ship by an incumbent.

Focus: Navel-Gazing-Centric vs Competition-Centric vs User-Centric

There are roughly three forms of focus for a company. Some large incumbents become so focused gazing at their own navels (or, relatedly their regulators’ navels) that they forget about users and start to create competitive openings for others to exploit. These large companies often have process after process and approval after approval on how to launch, how safe and “equitable” a product is, how to set goals and OKRs, performance reviews, executive reviews, legal reviews, marketing reviews, PR reviews, pre-review meeting reviews, how to map across every single geography at once, etc that they never launch anything that good anymore (or perhaps, launch at all). If you see a company that has not launched anything new in a while, they are probably doing the above.

Another form of company focus is competitor-centric focus. These companies continuously copy what their competitors are doing or have announced, versus what users are actually asking for or want. These companies are externally focused (which is good) but not customer focused (which is frequently bad), and often forget they exist to serve their customers. Sometimes these customers win in a market via lobbying or regulation, but often they end up derivative or staid. Some incumbents may succeed via this strategy - they just copy competitors or buy and bundle them, and keep relevance and market position via that motion.

There is an odd form of competition-centrism with a subset of early stage founders - where all the founder talks about is their competitors - often in a bitter tone about how X company is getting too much press and is raising money too easily and they do not deserve it - and maybe they too should go and do a giant unneeded funding round. For some reason, most bitter founder companies tend to fail.

Finally, there is user centric focus. User-centric companies tend to have a better understanding of their customer and their ongoing needs leading to superior product, sales, customer success, and pricing approaches. By aligning against the customer, more things tend to go right. While being customer-centric helps enormously in most cases, occasionally an incumbent is just too dominant for a superior customer experience and product to win.

What does this all mean for the “wrapper on OpenAI”?

The takeaway is that serving a customer need well is often more important (and harder) to think about than defensibility. In many cases defensibility emerges over time - particularly if you build out a proprietary data set or become an ingrained workflow, or create defensibility via sales or other moats.

The less building and expansion of the product you do after launch, the more vulnerable you will be to other startups or incumbents eventually coming after and commoditizing you. Pace of execution and ongoing shipping post v1 matters a lot to building one forms of defensibility above. Obviously, if your company is defensible up-front it is better then if it isn’t.

Beware the fast-acting incumbent[2] with the easy to add product space that is an obvious and necessary extension of what they are already doing.

NOTES

[1] Examples where things are less clonable includes areas that have deal or licensing moats (e.g. it takes a year to get a banking deal or a broker/deal license), IP (this is stronger in biotech where IP protection matters more), pre-existing open source software that is being commercialized by the team who created it, or other areas.

[2] In general founder-led companies will be more aggressive then non-founder led ones, although Nadella at Microsoft is an impressive counter example of this. Looking at a past history of bias to action (or not) is often revealing as well.

MY BOOK
You can order the High Growth Handbook here. Or read it online for free.

OTHER POSTS

Firesides & Podcasts

Markets:

Startup life

A brief guide to startup pivots

4 Ways Startups Fail

Founder Investors and Scout Programs

Co-Founders

Raising Money

Old Crypto Stuff: