Key Takeaways

  • Scott Wilson oversaw $17 billion in gains at Washington University in St. Louis, taking the endowment from under $8 billion to nearly $20 billion after cutting 36 hedge funds.
  • Wilson predicts half of today's venture-backed AI companies will not exist in ten years due to a recurring five-year cycle of private market excess.
  • The top five frontier large language models produce outputs so similar that users cannot tell them apart for 99% of tasks if placed behind identical interfaces.
  • The stall in venture capital distributions comes from public markets rejecting inflated private valuations rather than an issue with market liquidity.

The Commoditization Trap in Frontier Models

Scott Wilson runs nearly $20 billion for Washington University in St. Louis. Over nine years, his team generated $17 billion in investment gains, partially by cutting all 36 hedge funds in the portfolio and backing direct winners like SpaceX. When he looks at current artificial intelligence companies, he sees an old pattern repeating. “We're seeing the same kind of stupidity that we saw in all those other cycles,” (play from 25:34) Wilson noted, “and it's like every 5 years particularly in private markets.” (play from 25:39)

The core problem lies in product differentiation. Founders are spending tens of millions of dollars training or fine-tuning models that output nearly identical answers. Wilson highlighted this flaw directly: “If you take the top five kind of large language models and they all had the same user interface for 99% of problems or tasks, like you probably couldn't tell the difference.” (play from 28:22)

When buyers cannot spot a difference between vendors, their loyalty evaporates. Pricing power collapses immediately when open-source weights or subsidized alternatives show up. As Wilson observed, “When there's a free alternative like customer willingness to pay drops very very quickly.” (play from 29:06) If your core product relies entirely on model intelligence that five competitors can replicate by next quarter, your gross margins will compress to zero. Wilson expects the fallout will be severe: “I'm sure that half of these kind of I guess super intelligence is the correct term now, but these AI related companies like probably half of these companies don't exist 10 years from now.” (play from 28:01)

Why Public Markets Refuse Private Valuations

Venture capital investors frequently blame quiet initial public offering windows on frozen public markets. Wilson disagrees. Capital exists in public equity markets, but public fund managers refuse to purchase private assets marked at unrealistic multiples.

Private investors can trade shares between themselves at elevated marks during hype cycles, but those marks run into arithmetic at the exit door. Public market investors demand clear profit pools, defensible moats, and durable margins. A model wrapper trading at extreme forward revenue multiples with thin margins due to cloud compute bills cannot survive an initial public offering.

This valuation mismatch explains why venture capital distributions to limited partners have ground to a halt. Private valuations expanded far beyond any economic reality public buyers will accept. When public equity managers decline to buy overpriced shares, private portfolios freeze up. Startups that raised capital on inflated terms find themselves trapped, unable to raise flat rounds and incapable of going public without taking brutal write-downs.

What to Do With This

Review your product roadmap this week and list every feature where your edge is purely the intelligence of the underlying model. If a competitor can clone your core workflow simply by calling an updated API from Anthropic or OpenAI, delete it from your sales pitch. Shift your engineering focus toward building proprietary data loops, direct operational integrations, or local system permissions that an API upgrade cannot replace.