Core context
The power law governing venture capital returns has always been extreme — but AI is making it structurally more so, and most institutional allocators remain dangerously underexposed. That is the central argument of this conversation between A16z's David George and Jim Ka, and Accolade Partners' Aram Verdian. Their data point is striking: of 3,000 U.S. venture firms, only 20 have delivered consistent 3x net returns over two decades. Missing those 20 firms means settling for average venture returns of 1-2x net — worse than private equity, worse than public markets, with a 10-year lockup attached.
The AI-specific twist is that capital allocation now compounds competitive advantage in a way it never could before. Throwing money at a traditional startup created coordination failures. Throwing money at a frontier AI lab buys compute, and compute directly improves the product. This means economies of scale — previously the domain of network-effect consumer businesses — now apply to early-stage technical companies. OpenAI, Anthropic, and SpaceX alone represent $3.5-5 trillion in potential enterprise value, yet most LPs had minimal exposure before SpaceX's public listing.
The TAM argument is equally forceful. AI reached $100 billion in revenue in four years; SaaS took fifteen. But the more important point is that AI is not competing for software budgets — it is competing for labor budgets, which are 40 times larger in the U.S. economy. Healthcare IT spending runs $60-100 billion annually; the administrative and clinical labor AI can address in healthcare alone is a trillion-dollar market.
For allocators, the actionable framework is access, selection, and sizing. Access to the top 20 firms is necessary but insufficient — position sizing is where returns are actually made or lost. A late-stage fund that cannot put 5-10% of its capital into a single category-defining company cannot generate fund-returning outcomes. The risks are real: timeline to liquidity extends beyond a decade for most unicorns, and the pre-2022 SaaS vintage in private equity — assets bought at 25-32x EBITDA now trading at 2x revenue — illustrates how brutally the market punishes non-AI-native software.
20 / 3,000U.S. VC firms said to have sustained 3x net over 20 years · source-reported
$3.5–5TEstimated combined potential value of OpenAI, Anthropic, and SpaceX · source-reported
40×Claimed ratio of U.S. labor spending to software spending · source-reported