Bill Maris: How Google Could Crush AI Competitors, Why Small Funds Win, and AI's Atari Stage

Primary speaker: Bill Maris, who identifies himself as founder of Section 32, former founder/CEO of Google Ventures, and former Google Vice President of Special Projects. He says he incubated Waymo, Google X, Calico and other projects, and previously founded a web-hosting/data-center company.

Format: The source begins with introductory/promotional clips, moves into a prepared Bill Maris presentation built around four lessons, then shifts into a panel discussion with several hosts or participants.

Other speakers: Most panel turns are not reliably speaker-labeled. The transcript later explicitly addresses speakers rendered as “Sax” and “Freeburg”; those spellings/names should be verified before publication. A speaker named Thomas/Tom is repeatedly referenced but not fully identified in this transcript.

Transcript quality: The file contains heavy automatic-caption duplication, repeated fragments, stutters and music markers. These have been cleaned below without removing substantive content. Some slide-dependent information is missing because Maris says things such as “looked like this” while referring to graphics whose numerical contents are not captured. Several names and phrases are uncertain and are flagged accordingly. The opening also contains a historical-sounding clip saying Maris had raised $150 million for Section 32.

Verification status: This document records what the speakers said. No claim below should be treated as independently verified.

The introductory material presents Bill Maris as returning to investing after previously saying he was out. It calls him the founding CEO of Google Ventures and says he had raised $150 million for a new fund called Section 32.

In the opening clips, Maris argues that a smaller fund gives him an advantage because he can be more selective about both the companies he backs and the people he hires.

He says the fund's purpose is financial return and rejects alternative success metrics as effectively unmeasurable.

A separate introductory statement argues that the world has changed enormously over the previous century and predicts that AI will make the next century change by “orders of magnitude.”

Maris introduces himself as:

• founder of Section 32;
• former founder and CEO of Google Ventures;
• former Google Vice President of Special Projects;
• someone involved in incubating Waymo, Google X, Calico and other projects;
• and, before Google, founder of a web-hosting/data-center company.

He says the presentation will cover four lessons drawn from those experiences, beginning in 1997 when he had recently graduated from college with a neuroscience degree and unexpectedly found himself working on Wall Street.

The precise wording of lesson one is not cleanly captured in the transcript. The first story, however, concerns recognizing an important technological future before it becomes obvious.

Maris says he disliked his Wall Street job, including wearing a suit and commuting through the heat.

The useful thing that came from the job was discovering a server sitting underneath the jackets in an office closet. When he asked what it was, colleagues explained that the company's email and websites lived on it.

That produced what he describes as a moment of inspiration. His reasoning was simple: if one office could keep its website and email on a server in its closet, then perhaps he could put many companies' websites and email systems in his own closet and turn that into a business.

He says he believed he had “glimpsed the future” and immediately quit his job.

Maris says the company initially operated from his apartment in Vermont and was financed with credit cards.

The operation began with three servers — small, medium and large — and eventually grew to five.

One room contained the servers while another was used for work. The machines generated so much heat that he opened the windows despite Vermont's cold weather. He says the room then became cold enough that a glass of water on a desk could freeze by noon.

The apartment was also his home. He says what appeared in a photograph to be a blanket on his bed was actually a rug from Home Depot that he used for warmth.

During a thunderstorm, the apartment roof began leaking.

Maris knew water and servers were a dangerous combination. His landlord reportedly responded to the complaint by saying that roof leaks sometimes happened.

Maris therefore went to Home Depot, bought tar and a mop, climbed onto the roof during lightning and rain, and attempted the repair himself.

He did not know that he should tar from the far side toward the exit. Instead he worked in the opposite direction and literally tarred himself into a corner.

He characterizes the resulting choice as either allowing the servers to be electrocuted or risking being electrocuted himself. He says that, as an entrepreneur, he accepted the personal risk and survived; he jokes that his shoes remain stuck on the Vermont roof.

Maris explicitly introduces lesson two:

“To see the future, sometimes you need to be a little bit insane.”

His point is not merely that founders take risks. He argues that someone acting on knowledge about a future others do not yet believe can look irrational to everyone around them — like someone tarring a roof during a thunderstorm.

He illustrates the argument using slides provided by his friend Stewart Butterfield.

The slides compare presidential inauguration crowds:

• in 1989, only an occasional person appears to be taking a photograph;
• in 2005, the scene is not dramatically different;
• four years later, nearly everyone appears to have a camera.

Maris notes that this period coincided with cameras increasingly becoming integrated with mobile phones.

But the person he considers most interesting in the later photograph is someone apparently recording or livestreaming the inauguration using a laptop. Maris argues that this person probably appeared ridiculous at the time but had understood something about the future earlier than the crowd.

That becomes one of Maris's criteria for entrepreneurs: he looks for people who know or believe “a secret about the future” that most other people do not believe.

Maris jumps forward to 2007, when he was at Google.

He says Google needed a venture fund. The company had already started making investments, but there was no coherent strategy and no established budget.

Maris recruited Rich Miner, whom he identifies as a co-founder of Android, to help develop what Google Ventures could become.

They traveled around Sand Hill Road and spoke with anyone in the venture industry willing to talk, attempting to learn as much as possible.

Their eventual plan began by collecting as much historical venture-capital data as they could obtain. Maris says Google's position gave them access to a very large amount of data.

Their second step was using AI on the venture data.

Maris says that, at the time, Google would not allow them to describe the work as AI. He recalls being told that AI was science fiction, perhaps a century away if it ever happened, and that the phrase frightened people.

They therefore called it machine learning.

According to Maris, they used it for two main purposes:

1. designing an ideal venture portfolio by running millions of simulations, backtests and other data-science analyses;
2. determining the ideal fund size.

He says the approach generated outside attention, including a TechCrunch headline shown on a slide, and also attracted internal attention at Google. The actual headline and internal executive statement displayed on the slides are not captured clearly in the transcript.

Maris says the plan appeared crazy at the time but argues that the results justified it.

For the period 2009–2018, he compares Google Ventures with top-quartile and top-decile venture returns using slides.

He stresses that he is relying only on publicly available information, not proprietary Google data, and gives an estimated Google Ventures return of approximately 4.1x.

He then says investments he personally led followed the strategy more closely and performed at a level shown on another slide, but the actual numerical result is absent from the transcript.

This leads to lesson three: “don't bet against computer science.”

Maris says he has repeatedly seen appropriate computer science applied at the right time to the right problem produce the right answers. Even where the approach initially looks unreasonable, he says he would not bet against it.

Maris moves to 2017, when he decided to start his own investment fund.

People around him initially told him leaving Google was insane because he was giving up what he jokingly calls Google's “warm womb,” including free food and massages.

Once the idea became accepted, the advice changed: people told him to raise as much capital as possible because a larger fund would generate a large management fee.

Maris says he considered that advice against what he had learned previously and deliberately rejected it.

Maris says Section 32 has had six funds, investing in companies including CrowdStrike, Cohere and Coinbase.

He says the six funds average roughly $400 million each and claims all are performing in their top decile.

Where sufficient data exists, he says the only venture-performance measure he personally considers meaningful is DPI — distributed capital returned to investors rather than unrealized paper valuations.

Maris describes this conclusion as potentially heretical:

“Small funds outperform large funds.”

He presents it as a mathematical conclusion rather than merely his personal preference.

He also gives operational reasons. Smaller funds can be more focused. Maris says he has previously managed a multibillion-dollar fund with hundreds of employees and found the scale distracting because it prevented him from giving founders the amount of attention he wanted.

Maris gives the following figures for top-decile DPI performance:

• funds below $750 million: average return of 4.76x;
• funds larger than $1 billion: average return of 2.42x;
• funds below $750 million represented 95% of top-decile performers across the period he examined;
• he describes a sharp or “discontinuous” compression in returns above approximately $750 million.

These figures are speaker claims and require verification against the underlying dataset and period.

Maris illustrates the fund-size problem using a thought experiment.

For a $500 million fund, assuming the fund owns an average of 10% of portfolio companies:

• it requires approximately $5 billion of company exit value merely to return the fund's principal;
• to produce a 3x result, it requires approximately $15 billion of exit value.

He also states that the 75th percentile of venture loses money [wording/figure to verify] and says top-quartile performance has persistence.

He then applies similar arithmetic to a $7 billion fund.

The transcript says the fund would need approximately $210 billion in exit value to produce a 3x return under his assumed ownership economics. Maris says that figure exceeds total venture-backed M&A plus IPO exit value in most years.

He adds that the current year might be an exception.

As the panel begins, a host says he and Maris are old friends and that, when Maris started Google Ventures, the host's company was the first ex-Google company in which Maris invested.

The company is identified as Climate Corp, which the speaker says exited to Monsanto for $1 billion.

Maris does not remember the exact investment multiple but agrees that the investment performed very well.

The participants discuss whether Google Ventures entered the B, seed or A round; the exchange is somewhat confused by transcript overlap.

They then say they are going to “do it again with A Halo” [transcript unclear; company/name should be verified].

An unidentified host contrasts Maris's small-fund argument with information previously presented by Thomas.

The alternative thesis is that very large pools of capital can invest in later-stage, already-proven businesses where continued scaling is more predictable.

The host cites the idea that a company that has already reached approximately $1 billion might have a better probability of reaching $10 billion than a much smaller company has of reaching a comparable next stage.

He proposes a barbell strategy:

• small vehicles making traditional venture investments;
• very large vehicles investing in durable, proven late-stage companies;

with the possibility that both could generate roughly 3x returns.

Maris says he has not seen sufficient data science supporting the proposition that these extraordinary late-stage returns represent a durable long-term pattern.

He suggests they may instead reflect a peculiar moment in history involving companies headed toward extremely large, potentially multitrillion-dollar outcomes.

His second distinction is conceptual. An RIA or other asset manager accumulating large pools of assets is, in his view, performing a different business from venture capital.

Maris defines venture, as he practices it, as a craft built around concentrated commitments of both capital and personal time to entrepreneurs, including helping them build their businesses.

He explicitly says there is nothing wrong with late-stage investing; his point is that it is a different activity from the venture craft he is describing.

Maris then makes a separate objection to companies that present themselves using public-benefit or benefit-humanity language while keeping most early value creation limited to themselves and an elite group of private investors.

His argument is that if a company genuinely claims to exist for humanity's benefit, allowing the wider public to participate economically earlier may be more consistent with that mission.

He suggests such companies might therefore be better served by going public earlier rather than waiting until much of the appreciation has already occurred privately.

He summarizes the moral argument with the observation that “what humanity needs is money.”

Maris then offers a hypothetical, explicitly saying he does not speak for Google.

He asks what would happen to OpenAI and Anthropic's business models if Google dramatically reduced the price of AI tokens.

The initial wording around “cut the cost … to 80%” is malformed in the transcript. The follow-up makes the scenario clearer: a company could supposedly buy a basically comparable Gemini product for 80% less.

Maris asks why customers would not switch, arguing that such pricing would create severe margin compression and competitive pressure on rival AI companies.

When asked whether Google could actually pursue the strategy, Maris says:

“If I were Google, that's what I'd do.”

The hosts extend Maris's hypothetical.

They imagine Google using its financial resources to offer AI tokens at approximately 20 cents on the dollar, continually reducing prices whenever competitors respond.

One speaker describes this as:

“Capital as a weapon, tokens as a weapon.”

Another compares it with a strategy like Uber's: burn investor capital to capture market share and build a consumer and enterprise installed base.

The counterpoint is that, eventually, even such a strategy requires actual cash generation.

One unidentified speaker calls the scenario “100% possible” and “100% probable.”

Another cites a claim of approximately $1 trillion in spending commitments against $60 billion of revenue, then questions the idea of ultimately asking public investors to absorb the valuation. The exact company to which those figures refer is not stated clearly in this passage and must be verified.

The discussion broadens into whether companies remaining private for longer periods unfairly excludes ordinary investors from their most valuable growth phase.

A host frames the issue as affecting the vast majority of people who cannot participate in elite private markets.

The discussion specifically mentions:

• 401(k) retirement plans;
• S&P 500 inclusion;
• passive investment funds;
• ETFs.

A speaker claims some private-to-public companies are receiving unusual exceptions to S&P 500 rules and argues that passive funds may later be required to purchase those stocks only after enormous private-market appreciation has already occurred.

He acknowledges that some companies could still rise another 3x after going public, but questions where on the overall value-creation curve public investors are finally entering.

A host summarizes the critique in much stronger language: ordinary American 401(k) holders could be forced to buy overpriced assets after wealthy private investors received the early appreciation, turning retirement accounts into “bag holders.”

Maris does not explicitly adopt every element of that paraphrase.

His response is narrower: there is substantial risk in such a structure, but his central objection is to companies saying they are operating for the “benefit of humanity” while behaving in a way inconsistent with that rhetoric.

He says they could instead simply state that this is how they choose to operate their business and stop claiming a broader humanitarian purpose.

A host asks what happens to venture performance when extraordinary gains from today's largest companies are distributed.

The speaker predicts a handful of funds will generate almost unbelievable returns.

One example is rendered as “Founders two” [transcript unclear]. The speaker claims it could generate roughly $100 billion of return from $200 million of invested capital.

He also predicts funds from Maris's Google Ventures period could show enormous upside.

The important qualification is that these few enormous winners can make aggregate venture returns look spectacular even while the underlying distribution remains extremely bimodal:

• a small number of huge winners;
• many losing funds.

Maris responds that this is normal for venture and again says approximately 75% of funds lose money.

He then points out that a venture fund cannot convert a gigantic paper valuation into distributed cash unless someone else buys its shares.

A $100 billion paper gain therefore raises a basic question: who is the buyer?

Possible buyers include retail/public investors.

Maris says a public company ultimately has to support its value through a business case based on the discounted value of expected future cash flows.

He names SpaceX and Anthropic as examples whose eventual public-market valuations will have to survive that test.

He says the market may only reveal the answer after IPO lockups expire; he suggests some companies are experimenting with lockup structures that delay that moment.

Asked what he likes outside the small group of giant companies, Maris compares AI's evolution with the evolution of videogames.

He says people currently divide into AI doomers and utopians, but predicts reality will probably be neither as terrible nor as wonderful as the extremes claim.

He recalls playing Zork and Planetfall during the 1980s.

Those games were brittle text interfaces in which a player had to issue specific commands such as “grab the lamp,” sometimes failing because the computer expected another word such as “lantern.”

He compares this with today's most advanced consumer AI, arguing that the interaction model remains surprisingly similar in important ways. The slide showing the contemporary AI example is not identifiable from the transcript.

Maris argues that the transformation videogames experienced between the 1980s and today will occur in AI over only about the next five years.

Modern games became photorealistic, immersive and highly responsive. He expects AI to undergo an analogous transition much faster.

He expects problems analogous to old text adventures to disappear, including:

• weak or absent memory;
• inconsistency;
• session resets;
• brittle interaction.

He also expects ambient computing to emerge.

Maris says he does not plan to focus his investing on merely building bigger foundation models.

Returning to the videogame analogy, he argues that better videogames did not emerge simply because the stories became bigger. Progress required enabling systems such as:

• controllers;
• physics engines;
• GPUs;
• other underlying platforms and machinery.

Those enabling layers are what interest him in AI.

He describes the present moment as roughly the “Atari command line stage” of AI and predicts movement toward something analogous to “PlayStation 10” within five years.

The panel asks about Maris's previous life-sciences investing.

He says he founded Calico and has long been interested in the field. He mentions investments in Flatiron and a company rendered as “Veer” [transcript unclear], among others.

He likes healthcare partly because it can combine commercial success with improving people's lives.

However, he distinguishes broad healthcare and computational biology from therapeutic programs that require human clinical trials.

He says human-trial therapeutics are a specialist investment discipline on which his current organization is not spending much time.

He remains highly interested in computational biology.

An unidentified panelist says social media, particularly X, gives the impression of a renaissance in human health.

The speaker mentions topics including:

• purported pancreatic-cancer cures;
• cancer vaccines;
• peptides;
• an apparent explosion of new biomedical work;
• the increasing role of computation.

The same speaker explicitly says he does not know whether this impression is actually true.

He also observes that many of the major current biomedical winners do not appear to have been computationally driven; instead, they may reflect very strong conventional science done roughly a decade earlier.

Maris says he hopes computational biology will produce a major acceleration.

He recalls that when he created Calico, longevity research appeared comparatively fringe.

He says Section 32 is now an investor in NewLimit, which he identifies as a company associated with Blake Byers and Brian Armstrong, as well as other companies in the longevity field.

His point is that longevity no longer seems as strange or fringe as it once did.

Maris cautions that biology remains constrained by human physiology and FDA requirements.

Discovering a promising compound is, in his estimate, only about 5% of the work.

Researchers still have to determine dosing/titration and conduct extensive safety testing.

For this reason, he does not expect biomedical progress to become as immediately exponential as technology optimists might hope.

However, he says the trajectory could change if researchers achieve a sufficiently realistic in-silico simulation of a human cell.

Such a system could allow much more of biological experimentation to occur computationally and therefore accelerate development.

He says science is not there yet.

A panelist asks whether investment capital is moving toward India and China because biotechnology may reach market faster there than in the United States.

Maris's response begins with a missing subject in the transcript: [transcript unclear] “has always … indexed on human safety over speed to market.”

From context, the omitted subject appears to concern the U.S. regulatory approach, but this should not be silently repaired in publication.

Maris's actual principle is clear: prioritizing safety over speed creates costs, but moving too far toward speed can also cost lives. He says there is a balance to strike.

He adds that China and other countries are conducting research — including experiments in cloning and other areas — that, as far as he knows, is not occurring in the same way in the United States.

Maris then makes a politically sensitive claim.

He says the “gutting of the CDC and the NIH”, together with what he describes as an “anti-science vibe” in the United States, has moved intellectual attention elsewhere as funding for basic research dries up.

This is Maris's characterization and requires independent verification before publication.

Another panelist says China has developed its own “paper clip model” [meaning/context to verify] for recruiting top scientists from Europe and India, with scientists emigrating to China to work.

The speaker argues that this is talent the United States previously had access to and could recruit.

A participant says the U.S. needs to preserve its “neurological reserves”; another reformulates this as the country's brain trust.

The conversation then raises pressure on H-1B visa holders, with a speaker arguing that pushing skilled immigrants away makes other countries easier places to work and is harmful to American science.

A host asks whether “deep tech” — traditionally capital-intensive, risky and slow — is becoming more tractable because of AI, physics engines and related tools.

The host names Elon Musk's work with SpaceX and Tesla as an example of an entrepreneur successfully navigating this class of business.

Maris answers “Absolutely”, saying developments are moving much faster.

Asked what he would invest in, Maris says human biology and healthcare may be the largest total addressable market in the world and remain extremely interesting to him.

His other preferred areas are technologies underlying the AI revolution, including:

• physics engines;
• controllers;
• GPUs;
• other enabling machinery and infrastructure necessary to turn advanced AI concepts into operational reality.

Near the end, the moderator says he wants to bring “Sax and Freeburg” into the discussion.

He asks “Sax” about his venture-capital experience, referring to what sounds like Craft funds [transcript wording: “craft funds”].

The respondent says they have done four venture funds and two growth funds.

The moderator asks whether, on returning to the private sector, this speaker believes in the roughly $500 million venture-fund sweet spot advocated by Maris or thinks a firm has to become something more like Andreessen Horowitz.

The speaker says there is no need to become Andreessen Horowitz and argues that fund size determines fund strategy.

The speaker addressed as “Sax” explains that a fund generally needs to spread capital across roughly 20–25 portfolio companies to achieve portfolio construction/diversification.

Fund size divided across that number of companies determines typical check size.

Check size, in turn, determines which stage and part of the private market the fund can realistically play in.

Reflecting on Tom's presentation, “Sax” says he is considering whether investors might be better off concentrating on what used to be called late venture / early growth.

Under that approach, an investor might write approximately $50 million checks after companies have already demonstrated that they are breakouts, instead of competing in the much noisier earliest stages.

He presents this as a question rather than a settled conclusion.

Maris responds that the difficulty with the late-stage approach has to be understood through the incentives governing venture capital.

He gives three examples.

Maris says a $5 billion venture fund returning only 1.01x can still be described as approximately a 75th-percentile fund [claim to verify] and may therefore be able to raise another fund.

An institutional allocator such as the Stanford endowment may prefer an established giant vehicle because it can place hundreds of millions of dollars at once without creating career risk for the allocator.

The transcript renders one allocation figure as “two or 500 million” [unclear; likely $200 million or $500 million].

Maris says the GP managing a $5 billion fund that returns only 1.01x can personally make more money than Maris managing a $500 million fund returning 3x.

His point is that fund managers' personal economics can reward raising more capital even when the larger fund produces inferior returns for investors.

Maris imagines a researcher leaving OpenAI to form a startup.

His hypothetical offer:

• Bill invests $20 million;
• company valued at $100 million;
• Bill wants approximately 20%.

A giant fund, however, needs to deploy perhaps $250 million.

Instead of accepting the startup's $100 million valuation, the huge fund effectively says the company is now worth $4 billion and offers $250 million for a very small ownership percentage.

Maris argues that most founders will take the enormously higher valuation every time.

Only a seasoned entrepreneur who has already experienced the consequences of excessive early valuation might recognize the pitfalls and reject it.

Maris concludes that incentives are distorted across:

• institutional LPs;
• venture GPs;
• founders.

He predicts the pendulum will eventually swing back.

For that reason, he does not believe a strategy of remaining almost entirely late-stage and waiting to “snipe” established winners is a durable long-term venture model.

He says the historical data, in his view, suggests that strategy will not work indefinitely.

Exact quote: “We're going to invest for a financial return. Any other metric is impossible to measure and therefore won't succeed.”

Context: Introductory explanation of Section 32's approach.

Supports: Fund philosophy; measurable returns.

Exact quote: “The world's going to change by orders of magnitude.”

Context: Prediction about AI and the next century.

Supports: Long-term technology thesis.

Exact quote: “I think I've glimpsed the future.”

Context: His reaction after discovering a server hosting email and websites in an office closet.

Supports: Early technological insight; entrepreneurship.

Exact quote: “To see the future, sometimes you need to be a little bit insane.”

Context: Conclusion from the data-center and roof-repair story.

Supports: Contrarian founders; willingness to appear irrational.

Exact quote: “They know a secret about the future that most of us don't believe.”

Context: Maris describing the man recording an inauguration with a laptop before such behavior became normal.

Supports: Venture selection; contrarian insight.

Exact quote: “Don't bet against computer science.”

Context: His third stated lesson after discussing machine-learning-based venture portfolio design.

Supports: Data science; technology-driven investing.

Exact quote: “That's the only measure as far as I'm concerned in venture that counts is DPI.”

Context: His preference for realized distributions over paper performance.

Supports: Venture-performance measurement.

Exact quote: “Small funds outperform large funds.”

Context: Fourth lesson.

Supports: Central venture-capital thesis.

Exact quote: “It's distracting. You cannot give the attention to founders that I would like to give.”

Context: His experience managing a multibillion-dollar organization with hundreds of employees.

Supports: Operational argument for smaller funds.

Exact quote: “Well, what humanity needs is money.”

Context: Objection to companies claiming humanitarian missions while early financial gains remain limited to private insiders.

Supports: Private-market access; public-benefit rhetoric.

Exact quote: “If I were Google, that's what I'd do.”

Context: Asked about Google using low token prices against AI competitors.

Supports: AI price-war thesis.

Exact quote: “Capital as a weapon, tokens as a weapon.”

Context: Discussion of subsidizing AI token prices to capture market share.

Supports: AI economics; competitive pricing.

Exact quote: “My objection is don't say you're doing this for the benefit of humanity and do the other thing.”

Context: Clarifying his position after a host described retirement accounts becoming “bag holders.”

Supports: Difference between Maris's actual criticism and the host's stronger formulation.

Exact quote: “A handful of winners and a ton of losers.”

Context: Discussion of extraordinary venture returns masking an extremely bimodal industry.

Supports: Power-law venture economics.

Exact quote: “I think we're at the Atari command line stage of AI.”

Context: Comparing modern AI with the historical evolution of videogames.

Supports: AI infrastructure thesis; current technological maturity.

Exact quote: “We're going to get to the, you know, PlayStation 10 stage in the next five years.”

Context: Prediction that decades of videogame-interface progress will be compressed into a few years of AI development.

Supports: Five-year AI development forecast.

Exact quote: “That's like 5% of the work.”

Context: His estimate of how much of drug development is complete once a promising compound has been discovered.

Supports: Limits of computational acceleration in therapeutics.

Exact quote: “I think fund size determines fund strategy.”

Context: Discussion of whether venture firms should remain around $500 million or become giant asset managers.

Supports: Portfolio construction; venture strategy.

Exact quote: “The incentives are broken in all those ways, and the pendulum will swing back.”

Context: After explaining LP, GP and founder incentives favoring increasingly large funds and valuations.

Supports: Venture-market correction thesis.