Alphabet shares fell as much as 7.2% intraday on Monday, the worst session since February, after a second marquee AI researcher in a week confirmed he was leaving for a rival. On Wednesday, Noam Shazeer, Google’s vice president of engineering and co-lead of its Gemini models, announced he was returning to OpenAI. On Friday, John Jumper, the senior DeepMind scientist who shared the 2024 Nobel Prize with Demis Hassabis for AlphaFold, said he was decamping to Anthropic.
The market read the sequence as a narrative, not a coincidence.
It’s the second time Shazeer has walked out. He and Daniel De Freitas left Google in 2021 to found Character.AI, then came back in August 2024 as part of a reverse-acquihire dressed as a partnership. The return lasted less than two years. The man whose name sits on the transformer paper has now exited Google twice, and the second exit lands him at the lab that turned his architecture into a consumer category Google is still chasing.
Jumper’s departure cuts differently. AlphaFold has predicted over 200 million protein structures, and in his farewell post on X he marked “nearly 9 years” at DeepMind. Hassabis replied that the work “changed the world, and showed the field what was possible with AI for science and medicine.” Bloomberg reports Jumper was a key member of Google’s coding-tools team, and that a mood of frustration has settled over DeepMind, with employees describing no clear strategy for enterprise-facing AI coding tools.
That’s the gap the tape is pricing. Alphabet has raised $141 billion in debt and equity since October to fund the buildout. Weeks ago at I/O it unveiled the Gemini 3.5 Flash model and the Gemini Spark agent. And on Sunday, Microsoft CEO Satya Nadella told the Wall Street Journal the AI market was commoditized and called for less dependence on “AI Giants,” a framing that lands very differently when the giants are losing their headline scientists on consecutive weekdays.
Capex without retention is just capex. The institutional memory of how the transformer was built, and how AlphaFold was trained, is now distributed across at least three competing labs. Google still has the compute. It’s renting the talent back from itself, one rival at a time.