DeepMind Loses Shazeer and Jumper in a Single Week
The Transformer's co-inventor went to OpenAI and AlphaFold's Nobel laureate went to Anthropic — and the question is no longer whether Google can recruit talent, but whether it can keep it.
In the span of about a week, Google DeepMind lost the man who helped invent the architecture underneath every modern large language model and the man who won a Nobel Prize for the lab's most celebrated scientific achievement. Noam Shazeer is going to OpenAI. John Jumper is going to Anthropic. Either departure alone would have been a headline. Together, inside seven days, they read less like coincidence and more like a verdict.
The Transformer's co-author walks
Shazeer announced on June 18 that he is leaving Google to join OpenAI. His résumé is, in a literal sense, foundational: he is a co-author of the 2017 paper Attention Is All You Need, the work that introduced the Transformer — the architecture that backs ChatGPT, Gemini, Claude, and effectively the entire generative-AI industry. At Google he was a vice president of engineering and a co-lead of the Gemini models, the company's flagship effort to stay at the frontier.
What makes the exit sting is how recently and how expensively Google secured him. In August 2024, Google brought Shazeer back into the DeepMind fold through a reported ~$2.7 billion deal to license the technology of Character.AI, the startup he had left to found. The arrangement returned Shazeer and fellow researcher Daniel De Freitas to the company and installed Shazeer at the center of Gemini. Less than two years and several billion dollars later, he is leaving for the competitor Google most wants to beat.
Sam Altman framed the hire as a long-running ambition, posting that it was "only 10 years in the making" and calling Shazeer one of the people he has most wanted to work with since OpenAI's earliest days. For a company that already ships a frontier model line, the value is less about any single contribution than about denial: the person who co-designed the Transformer will now be improving OpenAI's models instead of Google's.
The Nobel laureate follows, in the other direction
Days later, John Jumper said he is leaving DeepMind after nearly nine years to join Anthropic. Jumper is the co-creator of AlphaFold, the system that predicts protein structures and that has been used to model more than 200 million of them — work that earned him a share of the 2024 Nobel Prize in Chemistry. He is, by any measure, the most decorated scientist the lab has produced, and the embodiment of DeepMind's "AI for science" identity.
Demis Hassabis responded publicly and graciously, thanking Jumper on X for an "extraordinary partnership" over nine years and crediting AlphaFold with changing the world. The warmth of the farewell did not change the substance: a Nobel laureate concluded that the most interesting place to do AI-for-science is no longer the lab that defined the genre.
That conclusion did not come from nowhere. Throughout 2026, Anthropic has been quietly assembling the apparatus for serious scientific work — opening wet labs, publishing research on agents built specifically for biological workflows, and announcing flagship partnerships with the Allen Institute and the Howard Hughes Medical Institute. Jumper is not joining a company that dabbles in science; he is joining one that has been building toward exactly the kind of program he ran.
Why two exits are a pattern, not noise
Individual researchers leave elite labs all the time; talent churn is the background hum of the industry. The reason these two departures landed differently is that they point in the same direction and away from the same place. Shazeer represents the architecture franchise. Jumper represents the science franchise. Losing both, to your two fiercest rivals, in the same week, invites the question that Fortune posed bluntly afterward: can DeepMind remain at the forefront of AI when its most symbolically important people keep deciding the frontier is elsewhere?
The mechanics of the AI talent market make this self-reinforcing. The marginal frontier researcher is scarce, the compensation is enormous, and the prestige of a destination compounds. When OpenAI lands a Transformer co-author and Anthropic lands a Nobel laureate in the same news cycle, both signals ripple outward to everyone deciding where to spend the next five years of their career. Recruiting is downstream of momentum, and momentum is a story people tell each other.
DeepMind is not a weak lab — far from it. Gemini remains genuinely competitive, the research output is deep, and Hassabis's bench is one of the strongest in the field. But the company now has to answer a harder question than "can we hire?" It has to answer "can we retain?" — and to do so against rivals who have just demonstrated, very publicly, that they can pull away even the people Google paid billions to bring home.
The frontier as a flow, not a place
The cleaner way to read the week is that the "frontier" is no longer a location you defend but a current you have to keep swimming in. Shazeer left a company that re-acquired him at staggering cost. Jumper left the lab where he did Nobel-winning work and received a heartfelt public send-off. Neither was pushed; both were pulled. The pull came from labs that have spent the last year making themselves the obvious next chapter — OpenAI by going full-stack and aggressive on hiring, Anthropic by building real scientific infrastructure rather than talking about it.
For Google, the consolation is that the talent did not vanish; it redistributed. For the field, the lesson is that no amount of history, prestige, or prior investment locks a researcher in place. The architecture that made all of this possible was, fittingly, about attention — and right now, the people who built the frontier are paying theirs to someone else.