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

2026-06-09 - 29 min - source - Read full transcript

Bill MarisDavid Sacks (host)David Friedberg (host)Chamath Palihapitiya (host)Jason Calacanis (host)

Key insights

Small venture funds structurally outperform large ones on DPI, and it isn't close.

Maris cites top-decile performance data: funds under $750M averaged 4.76x DPI versus 2.42x for funds over $1B, and sub-$750M funds represented 95% of all top-decile performers, with return compression accelerating sharply above the $750M line.

small-funds-outperform

The math of large funds doesn't work: exit value required to hit target returns exceeds total annual VC exit value.

At ~10% average ownership, a $500M fund needs $5B in exits just to return capital, and $15B to hit 3x. Scaled to a $7B fund, the same 3x target requires $210B in exit value, which Maris says exceeds total venture-backed M&A and IPO exit value in most years.

venture-fund-economics

Fund size mechanically dictates check size and market position.

Sacks frames it as fund size divided by roughly 20-25 portfolio names to hit diversification targets, which sets the check size a fund can write and therefore which stage and deal size it can compete for.

venture-fund-economics

GP incentives are misaligned with LP outcomes because fees scale with AUM, not performance.

Maris notes a $5B fund manager returning a mediocre 1.01x still earns more personally than a $500M fund manager returning 3x, because larger funds generate larger management fees regardless of DPI - and institutional allocators like endowments face no career risk for re-upping into large, safe-seeming funds.

venture-fund-economics

Google could use its balance sheet to wage a token price war that crushes OpenAI's and Anthropic's margins.

Maris argues that if Google cut Gemini token prices ~80% below cost, customers would switch to a functionally identical, far cheaper product, putting existential pressure on OpenAI's and Anthropic's business models; the hosts note this could be financed like Uber-style market-share buying rather than sustainable margin.

ai-price-war

Late-stage AI mega-rounds risk making public retail investors the exit liquidity for insiders.

The panel argues that keeping companies like OpenAI, Anthropic, and SpaceX private through enormous late-stage rounds - while getting exceptions to rules that let them into S&P 500-adjacent passive index flows - effectively forces 401(k) and ETF money to buy in only after most of the value has already been captured by early, concentrated investors.

ai-price-war

AI today is at its 'Atari stage,' not its mature form, and the gap will close fast.

Maris compares current AI to 1980s text-adventure games like Zork and Planetfall: brittle, turn-based, no persistent memory. He expects the equivalent of the leap from Atari to 'PlayStation 10' within five years, closing gaps like lack of memory, inconsistency, and session resets - compressing a decades-long gaming-industry maturation curve into a much shorter AI cycle.

ai-maturity-curve

The investable opportunity in AI right now is infrastructure and 'controllers,' not bigger foundation models.

Maris says he isn't investing in larger models themselves, drawing a parallel to gaming: better graphics engines, controllers, and GPUs made games better, not better writing. He's focused on the platform/infrastructure layer needed to make persistent, embodied AI real over the next five years.

ai-maturity-curve

Google Ventures' founding strategy validated using data science to systematize venture investing.

Maris and Rich Miner (Android co-founder) built GV's original strategy on aggregating historical venture data and running large-scale simulations to design portfolio construction and fund size - forced to call it 'machine learning' since Google executives considered 'AI' science fiction at the time. The strategy's estimated 4.1x return from 2009-2018 (using public data only) is Maris's proof point for lesson three: don't bet against computer science.

venture-fund-economics

Deep tech is becoming tractable for more entrepreneurs and investors because of AI enablement.

Maris says capital-intensive, long-cycle deep tech historically required someone like Elon Musk (SpaceX, Tesla) to pull off, but AI-driven physics engines and compute are now making the category more accessible to a broader set of builders and investors.

deep-tech-investing

Healthcare is the largest addressable market, but human-trial regulation caps how fast it can move even with AI.

Maris calls human biology/healthcare the largest TAM in the world and cites Section 32 investments in New Limit and past bets like Flatiron Health, but says finding a promising compound is only about 5% of the work - safety and clinical-trial titration remain the bottleneck - until AI can realistically simulate a human cell in silico, at which point he expects the pace to accelerate.

deep-tech-investing

U.S. science-funding cuts and immigration restrictions are pushing scientific talent toward China.

Maris says NIH/CDC funding cuts and an 'anti-science vibe,' combined with H-1B restrictions, are driving scientists out of the U.S. talent pool; he says China is actively recruiting top scientists from Europe and India with its own aggressive talent strategy, which he compares to a 'paperclip model.'

deep-tech-investing

Media referenced

Zork - other - Maris uses this 1980s text-adventure video game (turn-response gameplay, no memory between sessions) as an analogy for how primitive today's AI systems still are.
Planetfall - other - Named alongside Zork as another brittle 1980s text-adventure game Maris used to play, reinforcing the same 'AI is still early' analogy.

Companies

Google Ventures - Maris was founding CEO; built its early strategy on machine-learning-driven portfolio construction and backtesting, estimated at a 4.1x return from 2009-2018.
Section 32 - Maris's current venture fund, raised $150M for its latest vehicle; six funds averaging ~$400M each, all in the top decile by DPI.
Waymo - Incubated by Maris as Google's VP of Special Projects.
Google X - Incubated by Maris as Google's VP of Special Projects.
Calico - Longevity/life-sciences company Maris founded at Google; cited as an early, then-fringe bet on aging science.
CrowdStrike - Named as a Section 32 portfolio company.
Cohere - Named as a Section 32 portfolio company.
Coinbase - Named as a Section 32 portfolio company.
Climate Corp - Early Google Ventures investment; sold to Monsanto for roughly $1 billion, used as a case study in early-stage return multiples.
Monsanto - Acquired Climate Corp for about $1 billion.
Ohalo - Name uncertain (transcript renders it 'O'Hollo'); referenced as a follow-on agtech/genetics bet in the same vein as Climate Corp.
OpenAI - Discussed as vulnerable to margin compression if Google cuts Gemini token prices; also referenced via its valuation and IPO prospects.
Anthropic - Discussed alongside OpenAI as exposed to a Google-led token price war, and as an example of a late-stage private company whose eventual public listing will test whether retail will absorb its valuation.
Google - Discussed as able to use its balance sheet to cut Gemini token prices ~80% and squeeze competitors' margins in a price war.
SpaceX - Cited as one of the deep-tech companies proving the capital-intensive model can work, and as another late-stage private company whose public-market reception is uncertain.
Founders Fund - Cited as an example of a single fund that could print a ~$100 billion return on ~$200 million invested, illustrating venture's bimodal outcome distribution.
New Limit - Section 32 investment in longevity/computational biology, founded by Blake Byers and Brian Armstrong.
Flatiron Health - Named as a past life-sciences investment.
Andreessen Horowitz - Referenced as the model of a large, scaled venture firm that Sacks says he does not intend to become.
Craft Ventures - Sacks's fund; he says it has raised four venture funds and two growth funds.
Tesla - Cited alongside SpaceX as one of the few successful capital-intensive deep-tech companies, built by Elon Musk.

Techniques and frameworks

DPI (Distributions to Paid-In Capital) - Maris says this is the only fund-performance metric that matters in venture; used to argue small funds outperform ($750M threshold: 4.76x vs 2.42x for larger funds).
ML-driven portfolio construction and backtesting - Google Ventures' founding strategy: aggregate historical venture data and run large-scale simulations to design ideal portfolio construction and fund size, done under the label 'machine learning' because Google internally banned the term 'AI' at the time.
Fund-size-to-check-size math - Framework Sacks describes: fund size divided by 20-25 portfolio names determines check size, which in turn determines what stage/market a fund can compete in.

Summary

Bill Maris, founding CEO of Google Ventures and now founder of Section 32, opens with a solo keynote framed around four career lessons before joining the All-In besties for a Q&A. He starts with an origin story: quitting a Wall Street job after glimpsing a server closet in 1997, founding a bootstrapped web-hosting company out of his Vermont apartment (tarring a leaking roof mid-thunderstorm rather than risk the servers shorting out), and using that as a springboard into "you sometimes need to be a little insane to see the future." From there he recounts building Google Ventures in 2007 with Android co-founder Rich Miner, forced to call their data-driven, simulation-based investment strategy "machine learning" because Google leadership at the time considered "AI" science fiction. GV's strategy, by his estimate, returned about 4.1x from 2009-2018, which he uses as evidence for his third lesson: don't bet against computer science.

The talk's core argument, and its most concrete data, is about fund size. Maris presents top-decile DPI performance showing funds under $750M averaging 4.76x versus 2.42x for funds over $1B, with sub-$750M funds representing 95% of all top-decile performers. He walks through the underlying math: at typical ~10% ownership stakes, a $500M fund needs $5B in exits just to return capital and $15B to hit 3x, while a $7B fund scaled the same way would need $210B in exit value - more than total annual venture-backed exit value in most years. He also flags a structural incentive problem: GPs running huge funds earn more from fees on a 1.01x return than a small-fund manager earns on a 3x return, and institutional allocators face no career risk for re-upping into large, "safe" funds regardless of DPI.

The Q&A pivots into a sharp exchange about AI economics. Maris argues Google could use its balance sheet to cut Gemini token prices roughly 80%, functionally forcing a price war that squeezes OpenAI's and Anthropic's margins - "their margin is my opportunity" - since customers have little reason to pay more for a comparable product. The hosts (Sacks and Friedberg are directly named) push this further into a critique of late-stage AI mega-rounds: keeping companies like OpenAI, Anthropic, and SpaceX private through enormous rounds, while securing unusual exceptions into passive index flows, effectively makes retail 401(k) and ETF money the eventual buyer of value that's already been captured by early, concentrated investors - a dynamic Maris calls out directly as inconsistent with claims of building "for the benefit of humanity."

On where AI itself is headed, Maris frames today's systems as being at an "Atari command line stage," comparing them to brittle 1980s text-adventure games like Zork and Planetfall, and expects an equivalent leap to "PlayStation 10" within five years as problems like lack of persistent memory and session resets get solved. Consistent with that view, he says he isn't investing in bigger foundation models themselves but in the infrastructure layer beneath them - "controllers, physics engines, and GPUs" - the same category of investment that made games better rather than just better-written.

The conversation closes on deep tech and life sciences, where Maris calls healthcare the largest TAM in the world but cautions that U.S. clinical-trial safety requirements mean progress will stay incremental until AI can realistically simulate a human cell in silico. He also warns that U.S. funding cuts to the NIH and CDC, combined with H-1B restrictions, are pushing scientific talent toward China, which he says is aggressively recruiting researchers from Europe and India. Sacks closes out the session fielding a question about his own venture plans, agreeing with Maris's fund-size framework (fund size divided by roughly 20-25 names sets check size and market position) without committing to a specific fund-size target himself.

Notable Quotes

"This will be heresy to some, but small funds outperform large funds. This is simply the math." - Bill Maris

"I think we're at the Atari command line stage of AI, and we're going to get to the PlayStation 10 stage in the next five years." - Bill Maris

"Their margin is my opportunity. I'm going to give tokens out twenty cents on the dollar." - Bill Maris

"Don't say you're doing this for the benefit of humanity and do the other thing." - Bill Maris

"A five billion dollar venture fund that returns 1.01x gets to say that they earn the seventy-fifth percentile and can raise their next fund." - Bill Maris