Scott Wilson knows what it is like to make a contrarian bet that pays off big. His early investment in SpaceX helped create a multi-billion dollar windfall for Washington University’s endowment.
Now he’s making another one: The trillion-dollar AI labs are in major trouble.
Wilson argues that OpenAI and Anthropic have committed to enormous spending just as cheaper Chinese models are catching up. It’s a view that puts him squarely at odds with AI bulls like Vinod Khosla, who believes those massive infrastructure investments are precisely what will make the frontier labs hard to beat.
“These trillion-dollar-plus frontier companies are not worth the liabilities that they signed up for,” Wilson said. “There’s going to be a ton of free alternatives.”
OpenAI has returned huge markups for VC firms as well as schools such as the University of Michigan, whose early $20 million OpenAI investment is now worth over $2 billion. Wilson said he also had the chance to be an early investor in OpenAI but passed because he was not convinced the company would ever make money. Since then, his bearish view has hardened.
“When we looked at the original OpenAI deal that some of our peers participated in and made tons of money on paper, we were highly skeptical,” Wilson said. “I’ve gotten more skeptical over time.”
Wilson shared what he called his “unpopular opinion” onstage last week at Rock Yard Roundup, a Fort Worth gathering of roughly 100 asset managers, VCs, and tech founders. Between a honky-tonk crawl and a visit to the Stockyards Rodeo, he cast the frontier AI race as its own kind of Wild West: a frenzy of runaway spending and valuations that, in his view, is unlikely to end well.
Wilson’s warning comes as Chinese model makers such as DeepSeek, Alibaba’s Qwen, Zhipu AI, and Tencent have narrowed the gap with U.S. frontier labs on performance while offering major cost savings. That has fueled a broader investor debate over whether OpenAI and Anthropic, which is expected to go public next month, can sustain the enormous spending required to stay ahead if customers can switch to cheaper alternatives that are almost as good.
Ben Bergman/Business Insider
Wilson explained that he has become more dubious about frontier labs after talking to his colleagues on the ground in China, who have seen rapid progress in open-weight models. He grew more concerned after speaking with the founders of companies Washington University has invested in, who told him they were all switching to cheaper models.
“Whenever we talk to our portfolio companies, especially the ones who are heavy of AI, they are all moving aggressively towards open source,” he said. “It’s like any other high-cost U.S. good that has to compete with a low-cost import, particularly from China.”
Khosla says he’s never been more bullish on OpenAI
Data from OpenRouter, a platform where developers route requests among AI models, supports Wilson’s argument, though it is a limited snapshot. DeepSeek accounts for 25.3% of text-model requests on OpenRouter compared with 18.6% for OpenAI and 2.9% for Anthropic.
Vinod Khosla, a billionaire venture investor who was an early investor in OpenAI, told Business Insider he strongly disagrees with Wilson and said he has never been more bullish on OpenAI.
“People like that are silly, and they don’t understand how this works,” he said when told of Wilson’s view. “They have this notion that the model is the value.”
Khosla argued that the real advantage lies in controlling more of the expensive infrastructure beneath the model. A closed-model company can co-design chips, as OpenAI has done with its Jalapeño inference chip, around its own models and serving systems. That could reduce its reliance on Nvidia hardware and third-party cloud services, he said, lowering its cost to serve customers relative to an open-weight model run on someone else’s cloud.
“I’m not talking price, I’m talking about cost,” he said. “From power to data center to chips, to infrastructure software to inference models, the cost of the stack is almost certainly going to be lower in closed-source models than open-source.”
Do you have a tip about AI labs? Reach out to chief correspondent Ben Bergman securely on Signal at @benbergman.11
