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

What It Covers

Bill Maris, founder of Google Ventures and Section 32, shares four lessons from building a $150M fund, arguing that small funds under $750M structurally outperform large ones, AI remains at an early "Atari stage," and Google holds the power to crush competitors through aggressive token price cuts.

Key Questions Answered

• •Small Fund Math: Funds under $750M return an average 4.76x versus 2.42x for funds over $1B, and represent 95% of top decile performers. A $7B fund mathematically requires $210B in exits to hit 3x — exceeding total annual venture-backed IPO and M&A exit value in most years, making large funds structurally disadvantaged.
• •Google's AI Weapon: If Google cuts token pricing by 80%, OpenAI and Anthropic face existential margin compression. Enterprises would migrate to cheaper, functionally identical Gemini products, collapsing competitor revenue models. Maris argues this is the rational move for Google, using capital as a weapon to capture enterprise and consumer install base.
• •AI Infrastructure Over Models: Rather than betting on larger foundation models, invest in the picks-and-shovels layer — physics engines, controllers, memory systems, and session-consistency platforms. Just as better games required GPUs and controllers rather than better storylines, AI's leap from Atari to PlayStation requires platform infrastructure, not just model scaling.
• •Broken VC Incentive Structure: A $5B fund returning 1.01x lands in the 75th percentile and still raises its next fund, while the GP earns more in fees than a $500M fund returning 3x. This misalignment inflates valuations — large funds offer $250M at $4B valuations to deploy capital, distorting early-stage price discovery for founders.
• •Biotech Acceleration Ceiling: Computational biology is the tractable life sciences bet, not therapeutic drug development requiring human clinical trials. Even with AI-accelerated compound discovery, FDA safety requirements mean lab breakthroughs represent roughly 5% of total work. Full acceleration requires a realistic in-silico human cell simulation, which Maris says has not yet been achieved.