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

Summary

• After Google Ventures, Bill Maris returned to investing with Section 32, which he introduced as having raised $150 million. He says a smaller fund lets him be selective about companies and hires, while financial return is the only measurable objective; his central venture call is that funds below $750 million structurally outperform megafunds. He cites top-decile DPI of 4.76x below $750 million versus 2.42x above $1 billion; smaller funds produced 95% of top-decile performers, and DPI is, to the extent it can be measured, the only venture metric that counts in his view.
• Fund-size arithmetic makes the megafund hurdle almost market-sized just to return capital. At 10% ownership, a $500 million fund needs $5 billion of exits to return capital and $15 billion for 3x, while a $7 billion fund needs $210 billion—more than total venture-backed M&A and IPO exit value in most years. Maris says Section 32’s six funds average roughly $400 million and all perform in their top decile.
• Google’s war chest could turn tokens into a weapon and make current AI economics “go super critical.” Maris asks what happens if Gemini offers a basically identical product at 80% less: companies would have reason to switch, and OpenAI and Anthropic would face brutal compression. The companies may burn investor cash Uber-style to buy market share, but “at some point, you got to have cash generation.”
• The late-stage AI bounty remains paper wealth until somebody buys the stock, and Maris asks whether retail, 401(k), or passive capital will ultimately be that buyer. He objects to companies claiming public benefit while reserving early value for elite investors and later relying on exceptions that make passive funds and ETFs pick them up: “Don’t say you’re doing this for the benefit of humanity and do the other thing.” A putative $100 billion gain still requires a public-market buyer to justify the valuation through discounted future cash flows; lockups may delay that verdict.
• AI is only at the “Atari command-line stage,” so Maris would fund enabling layers rather than another large model. Using Zork’s brittle commands as the analogy, he expects gaming’s leap to photorealistic immersion to compress into roughly five years for AI, reaching the “PlayStation 10 stage.” The opportunities are memory, consistency, ambient computing, controllers, physics engines, GPUs, and other machinery.
• Computational biology could unlock healthcare’s enormous TAM, but biology and regulation keep the curve from becoming instantly exponential. Maris is less involved in life sciences than before but remains interested in the area; discovering a compound is “like 5% of the work,” with titration, safety, and human trials remaining. A realistic in-silico simulation of a human cell could accelerate progress. He also warns that gutting CDC and NIH support, an “anti-science vibe,” and pressure on H-1B holders are pushing scientific mindshare and people elsewhere.
• A panelist argues that the venture incentive stack rewards asset gathering even when fund returns are mediocre. A $5 billion fund returning 1.01x can claim 75th-percentile status and its GP can out-earn a $500 million fund returning 3x, while giant checks inflate a researcher’s $100 million startup toward a $4 billion valuation. The panelist concludes that late-stage sniping is not durable and “the pendulum will swing back.”

Deep dive

After saying he was out, Bill Maris returned to investing with Section 32, a new fund introduced as having raised $150 million. He says a smaller fund lets him be selective about companies and hires, and that financial return is the only measurable objective; other metrics are impossible to measure and will not succeed.

1. Seeing the future often looks irrational before it looks obvious

• Maris’s first “keyhole” into the future came in 1997, when he found the office’s email and website server beneath employees’ jackets. He quit Wall Street, founded a web-hosting and data-center company with credit cards, and began with three servers in a freezing Vermont apartment, eventually reaching five.

• The best founder specimen is literal: during a thunderstorm, Maris climbed onto the leaking roof with tar and a mop, worked from the door toward the far corner, and trapped himself. “My shoes, though, are still stuck on that roof.”

• His broader test for entrepreneurs is whether they “know a secret about the future that most of us don’t believe.” The example was a man recording the 2009 inauguration on a laptop while everyone around him used cameras—an apparently insane behavior that anticipated the future.

• At Google Ventures, Maris and Android co-founder Rich Miner gathered venture data, ran millions of portfolio simulations, and estimated ideal fund construction and size using what Google required them to call “machine learning”—because “AI is science fiction.” Public information led them to estimate GV’s returns at about 4.1x from 2009 through 2018: “Don’t bet against computer science.”

2. Small funds win because scale hurts both the math and the incentives

• When Maris decided to start his own fund in 2017, conventional advice was to raise as much as possible and harvest a large management fee. He chose the opposite: Section 32’s six funds have averaged about $400 million, invested in companies including CrowdStrike, Cohere, and Coinbase, and, he says, all perform in their top decile.

• His evidence is DPI-centered: top-decile funds below $750 million averaged 4.76x, versus 2.42x above $1 billion. Funds below $750 million represented 95% of top-decile performers, with “discontinuous return compression” once fund size crossed that threshold.

• The arithmetic is unforgiving. With 10% average ownership, a $500 million fund needs $5 billion of exit value to return capital and $15 billion to reach 3x; a $7 billion fund needs $210 billion, exceeding total venture-backed M&A and IPO value in most years.

• A panelist then tested a late-venture/early-growth alternative built around $50 million checks waiting for breakouts, and described the incentive stack. A $5 billion fund returning 1.01x can still claim 75th-percentile status, while its GP can make more than a $500 million fund returning 3x. Giant funds also reprice a researcher’s $100 million startup toward $4 billion by offering $250 million. The panelist concludes that incentives are broken and “the pendulum will swing back.”

3. Cheap Gemini tokens could break private AI valuations

• A panelist’s countercase was a barbell: small vehicles fund early venture while enormous pools compound in proven late-stage winners. Maris has “not seen the data science” showing that strategy persists beyond today’s “weird moment” of prospective multitrillion-dollar exits; collecting assets through an RIA is not the concentrated craft he calls venture, though he says there is nothing wrong with late-stage investing itself.

• Maris’s conditional attack scenario is stark: if Google cuts token prices by 80% and Gemini remains basically identical, why would a company pay more? Compression on OpenAI and Anthropic would “go super critical.” In response to the margin framing, he allows that the companies may burn investor cash Uber-style to acquire consumers and enterprises.

• The panel’s broader concern is that the strategy eventually requires cash generation: “$1 trillion in spend commitments on $60 billion of revenue.” Maris says that may be possible and probable, but “at some point, you’ve got to have cash generation.”

• His deeper objection is distributive. Companies keep much of the appreciation private while invoking public benefit, then receive exceptions to S&P 500 rules that may require passive funds and ETFs to buy them. Maris asks whether retail is the buyer for a theoretical $100 billion venture gain and says 401(k)s cannot participate in such companies while they remain private. The eventual buyer would have to make a public-market case for SpaceX-, Anthropic-, or similar valuations through discounted future cash flows; lockups may delay the market’s verdict.

4. The investable AI opportunity sits beneath the models

• Maris compares current AI with 1980s text adventures such as Zork: brittle commands, missing memory, inconsistent responses, and session resets. He expects gaming’s journey from turn-response text to photorealistic immersion to occur in AI within roughly five years—moving from the “Atari command-line stage” to “PlayStation 10.”

• Bigger stories did not create better games by themselves; controllers, physics engines, and GPUs did. Accordingly, Maris does not plan to invest in larger models. He wants the platforms and machinery enabling ambient computing, durable memory, consistency, and the next phase of the AI cycle.

5. Computation accelerates biology, but cannot yet delete biology

• Maris says he is not as involved in life sciences as before but remains interested in the field’s dual ability to “do good” and “do well,” having founded Calico and backed companies including Flatiron, Veer, and New Limit. The discussion distinguishes therapeutics requiring human clinical trials—a specialist area the panel says it is not spending much time on—from computational biology, which interests him.

• Longevity once looked like fringe science, but its normalization does not remove execution risk. Finding a promising compound is “like 5% of the work”; titration, safety testing, human biology, and FDA requirements keep progress from becoming as exponential as investors might want.

• The major unlock would be “a realistic simulation of a human cell in silico,” which could materially accelerate experimentation. Maris also says AI is making deep tech more tractable because things are moving faster. His investment interests include human biology and healthcare—the largest TAM in his view—and the physics engines, controllers, GPUs, and related infrastructure underlying AI.

• He contrasts the FDA’s safety-over-speed approach with countries that accept risks that can cost lives, while noting research in China including cloning experiments. Maris says gutting CDC and NIH support, drying basic-research funding, and an “anti-science vibe” are driving mindshare elsewhere. A panelist adds that China is recruiting scientists from Europe and India; Maris agrees that the US is losing its neurological reserves and says pressure on H-1B holders makes it easier for people to leave.