Andrew Ho walked out of OpenAI on July 29 after just eight months at the company. By 5 a.m. the next morning, still awake, he was posting advice to his former colleagues on X that landed like a grenade in the middle of a tech selloff: take the money and run.
Ho, a researcher who joined OpenAI in late 2025, announced he’s founding a startup to sell specialized reinforcement learning datasets to frontier AI labs. But it was his candid assessment of OpenAI’s valuation that drew hundreds of thousands of eyeballs overnight.
“I would strongly recommend taking liquidity if you’re eligible for tender offers. It seems somewhat implausible that the valuation is going to, like, 2x after the IPO, but it does certainly seem plausible that it could go down by 50%.”
The dataset play
Ho’s new venture, which has no public name or disclosed funding as of July 30, will focus on producing high-quality datasets for reinforcement learning. Specifically, he’s targeting long-horizon scientific reasoning and biology applications, the kind of complex agentic tasks that require verifiable ground-truth grading.
While at OpenAI, Ho co-authored GeneBench-Pro, a project squarely in the biological reasoning space. He’s essentially taking his expertise and packaging it for the broader market of frontier AI labs that are all hungry for the same thing: better training data.
The OpenAI talent exodus
Ho’s departure isn’t happening in a vacuum. It’s part of a pattern of high-profile exits from OpenAI throughout 2026 that has started to look less like normal turnover and more like a slow-motion talent bleed.
Ho’s warning to colleagues about tender offers suggests he believes the current private-market pricing has gotten ahead of reality. If OpenAI’s valuation is already stretched at its current level, he thinks a 50% haircut is plausible — the kind of asymmetric risk profile that makes rational people sell.