Key Takeaways
- • Scott Wilson, chief investment officer of Washington University's endowment, warned that OpenAI and Anthropic face serious financial trouble because they have locked in massive spending obligations while lower-cost competitors close the performance gap.
- • Wilson said frontier AI companies valued at more than a trillion dollars "are not worth the liabilities that they signed up for," according to Business Insider, and predicted "a ton of free alternatives."
- • Wilson cited Chinese model makers including DeepSeek, Alibaba's Qwen, Zhipu AI, and Tencent as narrowing the gap with U.S. frontier labs at lower cost.
- • Vinod Khosla, an early OpenAI backer, rejected the thesis, arguing that companies owning their full stack from chips to models hold a durable cost and performance edge.
- • Anthropic is reportedly expected to go public next month, and both companies continue to expand infrastructure commitments, including Anthropic's cloud agreement with Nscale valued at roughly $45 billion.
Summary
Scott Wilson, chief investment officer of Washington University's endowment, is warning that OpenAI and Anthropic face serious financial trouble. He argues the companies have committed to large spending obligations just as lower-cost competitors rapidly catch up. Wilson shared what he called his "unpopular opinion" last week at Rock Yard Roundup, a Fort Worth gathering of roughly 100 asset managers, venture capitalists, and tech founders, according to Quartz's reporting of Business Insider coverage.
Wilson's concern centers on Chinese model makers, including DeepSeek, Alibaba's Qwen, Zhipu AI, and Tencent, which have narrowed the performance gap with U.S. frontier labs while offering lower costs. He said conversations with colleagues in China and with founders of Washington University portfolio companies deepened his skepticism. He noted that AI-heavy portfolio companies are moving aggressively toward open source, comparing the dynamic to "any other high-cost U.S. good that has to compete with a low-cost import." Wilson passed on an early opportunity to invest in OpenAI, having doubted the company could ever be profitable. The university's endowment did make an early bet on SpaceX, which produced a multi-billion dollar return for the fund.
Usage data from OpenRouter, a service that routes developer queries to various AI models, offers partial support for Wilson's view, though the figures cover only a narrow slice of overall usage. According to Business Insider, DeepSeek handles 25.3% of text-model traffic on the platform, OpenAI 18.6%, and Anthropic 2.9%.
Vinod Khosla, a venture investor and early OpenAI backer, pushed back, calling such views "silly." He argued that companies building and owning their full technology stack, from power and data centers to custom chips and inference models, hold a durable edge that open-source alternatives cannot easily replicate. OpenAI has developed its own Jalapeño inference chip. Khosla said the cost of the stack "is almost certainly going to be lower in closed-source models than open-source."
Both companies are expanding infrastructure commitments. Both have pursued data center deals across the U.K. and Nordics alongside large U.S. buildouts. Anthropic has reached a cloud agreement with Nscale valued at roughly $45 billion, while OpenAI has pledged several gigawatts of additional capacity through its Stargate initiative. Anthropic is expected to go public next month.
Wilson's concern centers on Chinese model makers, including DeepSeek, Alibaba's Qwen, Zhipu AI, and Tencent, which have narrowed the performance gap with U.S. frontier labs while offering lower costs. He said conversations with colleagues in China and with founders of Washington University portfolio companies deepened his skepticism. He noted that AI-heavy portfolio companies are moving aggressively toward open source, comparing the dynamic to "any other high-cost U.S. good that has to compete with a low-cost import." Wilson passed on an early opportunity to invest in OpenAI, having doubted the company could ever be profitable. The university's endowment did make an early bet on SpaceX, which produced a multi-billion dollar return for the fund.
Usage data from OpenRouter, a service that routes developer queries to various AI models, offers partial support for Wilson's view, though the figures cover only a narrow slice of overall usage. According to Business Insider, DeepSeek handles 25.3% of text-model traffic on the platform, OpenAI 18.6%, and Anthropic 2.9%.
Vinod Khosla, a venture investor and early OpenAI backer, pushed back, calling such views "silly." He argued that companies building and owning their full technology stack, from power and data centers to custom chips and inference models, hold a durable edge that open-source alternatives cannot easily replicate. OpenAI has developed its own Jalapeño inference chip. Khosla said the cost of the stack "is almost certainly going to be lower in closed-source models than open-source."
Both companies are expanding infrastructure commitments. Both have pursued data center deals across the U.K. and Nordics alongside large U.S. buildouts. Anthropic has reached a cloud agreement with Nscale valued at roughly $45 billion, while OpenAI has pledged several gigawatts of additional capacity through its Stargate initiative. Anthropic is expected to go public next month.
Go deeper than the headline
Research every firm in this story, track the ones that matter and get the daily briefing — free for 7 days.