OpenAI, Anthropic, Meta Hire 22 Professors From Top US Schools

The interesting thing about the latest wave of AI hires is not the size of the pay packages, it is where the people are coming from. The Atlantic reports that at least 22 professors from schools including Stanford, Berkeley, Harvard, MIT and Carnegie Mellon have left or taken leave in 2026 to join OpenAI, Anthropic, Meta or Google DeepMind. Anthropic recently picked up a Stanford economist and a philosopher from UT Austin, and OpenAI now employs top mathematicians and physicists.

What makes this different from a normal round of Silicon Valley recruiting is who is being recruited. These are not just senior engineers, they are the people who choose the research questions, teach the next cohort of scientists, and provide the outside scrutiny of systems that increasingly show up in banks, hospitals and government.

The mechanical reason is straightforward. Training a frontier model reportedly costs hundreds of millions of dollars in GPU time alone, and no university grant is going to compete with that. If your research agenda is about how a very large model behaves when you change the reward signal, the honest answer is that you can only actually run that experiment inside one of four companies. Compensation is doing work here too, but the compute gap is the load-bearing thing.

The reporting is careful on one point that matters. Most of these hires are formally on two-year leaves of absence rather than resignations, so whether this is a temporary detour or a permanent transfer of expertise is not settled. What the piece does not give you is the fine print of those contracts, whether the research these professors do inside the labs will ever be published, or what happens to the graduate students left mid-thesis by an advisor who is suddenly under an NDA.

The forward-looking piece is who benefits from the vacuum. If the top-five schools quietly stop being the center of frontier AI research, that opens real ground for second-tier universities, international labs, and independent nonprofits to become the places open work still happens. A smaller world of open AI research, not necessarily an empty one.