Only 20 of 3,000 VC firms hit consistent 3x returns

a16z's David George and Jen Kha sit with Accolade Partners' Aram Verdiyan to argue that AI has made venture's power law more extreme than at any point in the last 10 to 20 years, and that most institutional allocators are structurally locked out of the returns that matter. The conversation reframes AI not as software's next chapter but as a force hitting labor, healthcare, and $30 trillion of GDP at once.

• Capital as moat — David George says that for the first time in his career, throwing capital at a company (via compute) directly compounds its competitive advantage, unlike past eras where overhiring broke startups.
• $3.5-5T in three labs — George says SpaceX, OpenAI and Anthropic together represent roughly $3.5 to $5 trillion of potential enterprise value, exposure most LPs lacked before SpaceX went public.
• 100B in 4 years — Verdiyan says AI reached $100 billion in revenue in four years, versus 15 years for SaaS to hit the same mark.
• TAM beyond software — Verdiyan notes healthcare IT spend is only $60-100 billion a year, but AI can target the full trillion-dollar labor value of claims, billing and administration, making AI's TAM potentially 10x bigger than traditional SaaS.
• 20 of 3,000 firms — Verdiyan cites Accolade's data on 3,000 US VC firms showing only 20 have achieved consistent 3x net TVPI returns over three to four funds across two decades.
• Average VC return — He adds that Cambridge data shows the average venture return over the last 10 years is just 1 to 2x net, worse than private equity or public markets.
• Late-stage sizing math — George says fund-returning math now exists in late stage too: a top holding should be 5-10%+ of a late-stage fund, which requires an early-stage franchise to secure that allocation.
• Loss rates by stage — George says a16z's best early-stage funds run about a 60% loss rate, while growth-stage loss rates run 10-20%, which he calls appropriate risk-taking.
• AI spend dispersion — George cites data showing the median US company spends $12 per employee per month on AI, versus $7,000 per employee per month at the top 1% of companies surveyed.
• Traction confusion — Verdiyan describes companies going from zero to $5 million ARR-run-rate in a month with no renewal cycle, often selling within an accelerator cohort, versus rarer companies with genuine multi-million ARR like the next Cursor.
• Cursor deal doubted — George recalls people were still calling Cursor 'dead' the morning its roughly $300 million ARR, $400+ million round/acquisition by OpenAI was announced.
• Private equity squeeze — Verdiyan says 2021-22 vintage LBO software deals averaged 25-32x EBITDA with over $200 billion in debt taken out, and those companies are now worth roughly half that as public SaaS multiples compressed to about 15-20 companies trading above 10x revenue.
• CalPERS pivot — George cites CalPERS shifting its portfolio from 91% to 58% in traditional allocations and from 9% to 43% in venture and growth.
• Next frontier bets — George says the next major value creation will come from consumer AI beyond chatbots, robotics, autonomy (fewer than 10,000 Waymos live in the US), and healthcare, which is 18% of GDP but barely touched by AI.

In their words

We've looked at the data of 3,000 venture capital firms in the US. Only 20 have achieved consistent 3x net returns over the last two decades.

For the first time, you can take capital and throw it at a company and it compounds their advantage.

I can tell this for myself, I've been chronically wrong about how big these outcomes can get.6:28

If you look at Cambridge data, the average venture return over the last 10 years is 1 to 2x net. You'll do better in private equity. You'll definitely do better in the public markets.12:32

Disclosure · The speakers are a16z general partners and an LP at Accolade Partners discussing their own funds, portfolio companies (including Cursor, Harvey, Airtable/Intercom) and a16z's own late-stage and opportunities funds; this is effectively a promotional conversation for their investment strategy and fund products.

One thing to add — One thing to add — the striking "20 of 3,000 firms" and "$12 vs $7,000 per employee" statistics come from proprietary data the speakers themselves compiled, so they should be read as arguments for concentrating capital with firms like a16z rather than independently verified industry benchmarks. The panel's framing of AI as "not zero-sum" and "everything works" deserves scrutiny given both speakers have a direct financial interest in sustaining high valuations across the stack.</note> </invoke>

One thing to try tonight
Pull up your own company's (or a client's) monthly AI tool spend per employee and compare it against the $12 median and $7,000 top-1% figures cited in this talk to gauge where you actually sit on the adoption curve.