How I Invest with David Weisburd
E447: Anthropic at $2 Trillion? The AI Bubble, and Why the Best Startups Always Look Overvalued
26 Oct · 63 min
Key points
- Matthew Harris of Bain Capital Ventures argues AI's winners look overvalued at every round because extraordinary founders justify paying about 50% more on a multiple.
- Harris says the consensus market, where a leader is anointed and money piles in, differs from 2021's frenzy, and he disputes Mark Andreessen's claim that no diamonds in the rough remain.
- Harris says Anthropic's large legal and insurance customer base makes it unlikely to compete with those customers directly, while he calls compute the prime battlefield over the next five years.
Why the best startups always look overvalued, and what it means for AI
Harris argues that AI looks like a bubble partly because every great venture winner looks overvalued when it is funded. Paying up for extraordinary founders is how funds capture power law returns, but the same belief in winners bids prices higher and inflates bubbles. His firm responds by betting on rapid AI progress while pacing its investing at a steady $500m a year.
Takeaways
- Winning startups look overvalued when you buy them. Harris says extraordinary founders justify paying about 50% more on a multiple basis. Databricks, Stripe and Ramp looked pricey at every round yet became hugely valuable, so a rule of rejecting deals not 50% too expensive would miss the power law winners.
- Believing in winners is what inflates bubbles. When investors expect to back a power law company, each round can raise price by about 50%, and rival bidders keep pushing. Harris compares AI to the SaaS cycle that began in 2012 and 2013 and ended in November 2021, placing AI in the eighth inning.
- Bain Capital Ventures bets superintelligence arrives within one to three years. The firm formed this view about eighteen months ago. Two themes, robotics and AI native services such as the insurance claims business Reserve, explain 75% of its investing since. It has made 20 AI native services investments since late 2022.
- Frontier labs are constrained by their own customers. Harris argues Anthropic earns heavily from legal and insurance clients, including Travelers, Aon and Willis Towers Watson, so it will likely not compete with them. Google and Amazon dropped consumer finance plans for similar reasons, such as AWS serving JPMorgan Chase.
- Compute is the next big battlefield. Harris says access to compute may matter most over the next five years, as talent did in the prior five. The OpenAI and Anthropic race has shifted from artificial general intelligence toward revenue and profits, and he sees Musk as best at building data centers.
- A $2tn Anthropic valuation depends on continued progress. Harris calls it reasonable given revenue growth and shocking mainly because of its size. If AI only continues on its current trajectory, it is far too high. Liquidity pressure from sellers could hit prices, though tranche-based unlocks can soften it.
- Adoption is early, so this may be an investment bubble. Weisburd cites Lucas Swisher's point that about 50m engineers use most tokens. Harris says agents could create hundreds of billions of new entities and cheaper open weight models expand demand. He puts above 10% odds on depressed returns for this vintage.