Scott Wilson, the chief investment officer overseeing Washington University in St. Louis's endowment, is making a bearish case against two of the AI industry's most valuable companies. He argues that OpenAI and Anthropic have committed to enormous spending just as cheaper Chinese open-weight models are becoming credible substitutes for more work.
Business Insider reports that Wilson presented the view at a gathering of asset managers, venture investors and founders in Fort Worth. Wilson, whose role is confirmed by WashU's investment office, said conversations with colleagues in China and companies in the university's portfolio made him more skeptical of the frontier labs' economics.
The pressure is about cost, not only capability
Wilson's thesis is that customers do not always need the most capable model available. If an open-weight system is good enough for a task and materially cheaper, companies can switch without waiting for it to beat every frontier benchmark.
That pressure is visible beyond his comments. Axios reported that OpenAI, Anthropic and other providers are increasingly competing on price as model demand grows. Ars Technica reported that a Chinese open-weight model came within a point of a leading Anthropic model on one common software-task evaluation while costing far less per completed task. Those comparisons do not prove one model is best for every workload, but they help explain why buyers are testing alternatives.
Business Insider cites OpenRouter request data as limited support for Wilson's argument. DeepSeek accounted for a larger share of text-model requests on that routing platform than OpenAI or Anthropic at the time examined. OpenRouter is one developer marketplace, however, not a complete measure of enterprise AI demand.
Khosla disputes the economics
Venture investor Vinod Khosla told Business Insider that Wilson's conclusion overlooks the value of controlling the full infrastructure stack. His counterargument is that a closed-model company can design chips and optimize data centers, software and inference around its own systems, potentially lowering the actual cost of serving customers even if its listed prices remain higher.
The disagreement turns on whether frontier labs can reduce costs and retain enough performance-sensitive customers to justify their capital commitments. Wilson sees open-weight competition eroding that advantage; Khosla sees integrated infrastructure making it more durable. Neither position establishes what OpenAI or Anthropic will ultimately be worth, but together they frame the financial test created by rapidly improving, cheaper models.
Technology