Bill Maris’s core argument is that venture works when funds stay small, concentrated, and disciplined about DPI and cash returns, while AI is about to reshape everything from venture math to product categories. He says Google-era scale, late-stage megafunds, and “benefit humanity” branding can distort incentives, especially if public investors end up buying expensive AI shares after lockups. Across the conversation he points to Section 32’s smaller-fund model, Google’s possible price war in tokens, and a five-year AI shift from “Atari command line” to a much more capable platform stack. He is bullish on AI infrastructure, human biology, and deep tech, but bearish on giant late-stage strategies and on the idea that multi-trillion-dollar IPOs can be safely supported by public-market buyers.
Bill Maris said he is returning to investing after previously saying he was out, and that he raised $150 million for his new fund Section 32.
He said he founded Section 32, previously founded and ran Google Ventures, served as Google’s VP of Special Projects, and incubated Waymo, Google X, Calico, and other projects.
He recounted starting a web-hosting/data-center business in 1997 after seeing a server in a Wall Street office closet, quitting immediately, and later operating the business from his Vermont apartment with servers in one room and his bed warmed by a Home Depot rug.
He said the apartment roof leaked in a thunderstorm, so he went to Home Depot for tar and a mop, climbed onto the roof, tarred himself into a corner, and had to choose between the servers getting electrocuted or him getting electrocuted; he said he survived and his shoes are still stuck there.
Bill Maris said he looks for entrepreneurs who know a secret about the future that most people do not believe, and that seeing the future sometimes requires being “a little bit insane.”
He said Google Ventures was built after Google asked for a venture fund in 2007, he and Rich Miner learned from Sand Hill Road, gathered as much venture data as possible, and used machine learning to model portfolio construction and fund size.
He said publicly available information suggested Google Ventures returned about 4.1x, and that this reinforced his lesson to “not bet against computer science.”
He said Section 32 has had six funds averaging about $400 million each, invested in names including CrowdStrike (CRWD), Cohere (COH), and Coinbase (COIN), and that all six funds are performing in their top decile.
He argued that venture should be judged on DPI, and that small funds outperform large funds because they can be selective, focused, and more hands-on with founders.
Bill Maris cited data showing funds under $750 million averaged 4.76x in top-decile DPI performance, while funds above $1 billion averaged 2.42x, with funds below $750 million making up 95% of top-decile performers across that period.
He said the math gets brutal for large funds: a $500 million fund owning 10% of a company needs $5 billion of exits just to get money back, while a 3x outcome requires $15 billion; a $7 billion fund would need $210 billion of exit value for a 3x result.
He said Google could cut token prices to 80% off, or even to 20 cents on the dollar, and he called Google-like price cuts 100% possible and 100% probable because Google has a war chest and a money-printing machine.
He warned that if Google used price cuts to grab share, competitors like OpenAI and Anthropic would face compression and pressure, and that public retirement accounts could become the bag holders if overvalued AI companies end up being sold into the public markets.
He said the public-market test for companies like SpaceX or Anthropic will come about six months after lockups are worked through, when investors have to decide what discounted future cash flows are actually worth.
Bill Maris said he does not want to invest in larger models, but instead wants platform layers around AI — controllers, physics engines, GPUs, and other machinery — and he expects AI to move from an “Atari command line stage” to a “PlayStation 10 stage” within five years.
He said biotech is constrained by FDA timelines and human biology, so it will not become as exponential as some want, but that realistic in-silico human cell simulation could accelerate it; he also said he is focused on human biology and healthcare because he thinks it is probably the largest TAM in the world.