OpenAI researcher quits, warns colleagues equity may crash 50%

On Wednesday morning, Andrew Ho announced his departure from OpenAI after eight months as a researcher. He posted that he was starting a company to sell high-end reinforcement learning datasets to frontier AI labs. Hours later, still awake around 5 a.m., he posted unsolicited advice to his former colleagues: take liquidity in tender offers. "I would strongly recommend taking liquidity if you're eligible for tender offers," he wrote, adding that while a 2x valuation gain after IPO seemed implausible, a 50% decline was credible. The posts arrived during a tech selloff that had pushed the Nasdaq 100 into correction territory the previous day, drawing hundreds of thousands of anxious responses.

In an interview with Fortune, Ho explained his position. He holds about $700K in shares legally locked until after IPO and lockup period—shares he cannot sell despite his bearish view. "I'm just stuck," he said. Ho's concern is rooted in what analysts call a "Red Queen's race": frontier labs must pay more for each training cycle, yet the competitive advantage is temporary. Cheaper competitors like Moonshot's Kimi can close capability gaps for a fraction of the cost through distillation. Revenue will almost certainly rise with each new model, Ho said, but the question is whether it rises fast enough. "If you miscalculate even by just a very fine amount, that can be the difference between life or death for a company." Most peers believe they're on the cusp of recursive self-improvement (RSI), where AI systems improve themselves iteratively in a parabolic curve. Ho does not buy it. Research, he argued, isn't limited by model intelligence but by "research taste"—the ability to propose good experiments and recognize which results matter. "You can't reason your way to phenomena," he said.

Ho's skepticism puts him at odds with podcaster Dwarkesh Patel, who argued compute costs could rise 10x and thus revenue could scale 10x with RSI. But it aligns with Wall Street: Meta sold off 10% last Wednesday and Google 8% the previous week, both on fears that companies funding the buildout won't recoup their investment. Speaking as part investor, Ho identified the real winners: "Nvidia and Micron," which sell chips to everyone in the race. Labs have two hard paths out of the squeeze. They could build their own chips and break Nvidia's pricing power, or move into applications and products—like Meta buying Cursor—to capture more of the value their models create. "Right now," Ho said, "the labs capture a fraction of the value their models create in the economy, and that gap is very undercapitalized at the moment." Ho's first products will focus on long-horizon scientific reasoning and statistical analysis, filling the gap left after the low-hanging, easily-verifiable fruit has been picked. For now, he is fielding interest in his unnamed company while stressing over his locked-up equity: "It's this huge proportion of my net worth."