News · 2026-06-26
Google DeepMind loses four senior scientists in six days, including a Nobel laureate
In the span of about a week, Google DeepMind lost four senior researchers to its two fiercest rivals. Noam Shazeer, a co-author of the original Transformer paper and a co-lead on Google's Gemini models, announced he is going to OpenAI. John Jumper, the DeepMind scientist who shared a 2024 Nobel Prize for the protein-folding system AlphaFold, left for Anthropic within days, and two more Gemini and AlphaFold contributors, Jonas Adler and Alexander Pritzel, reportedly followed Jumper to Anthropic. Coverage has been collected at outlets including Let's Data Science and a market-reaction writeup at witho2.
Why this is a real story and not gossip: in AI, the people are the moat. The architecture behind almost every modern chatbot, the Transformer, came out of a single small team, and Shazeer was on it. AlphaFold, which Jumper led, is arguably the most important scientific application of deep learning to date. When that caliber of person changes employers, the knowledge, the instincts, and often the next breakthrough move with them. Losing one is bad luck. Losing four in a week reads like a signal.
What actually happened, and why. Three forces seem to be pulling at once. The first is money, but a specific kind. Both Anthropic and OpenAI are widely expected to go public within months, which means joining now means receiving pre-IPO equity, the sort of ground-floor stake that an established giant like Google simply cannot match. Anthropic has been raising at a valuation approaching a trillion dollars, which makes its stock options extraordinarily attractive. The second force is scientific opportunity: Jumper's move to Anthropic, just as that company deepens a push into biology and after it recently brought on other prominent researchers, looks like a bet on a more open frontier. The third, quieter force is internal friction, with reports pointing to compute allocation as a sore point, the idea that DeepMind's researchers get a smaller slice of Google's enormous computing fleet than its cloud and recommendation businesses do.
An analogy for the compute friction. Imagine the best chefs in the world working in a restaurant that also runs a massive catering operation. The catering side is reliable and profitable, so it gets first claim on the ovens. The chefs, who want to invent new dishes, keep finding the kitchen booked. Eventually a rival opens a restaurant built entirely around their cooking, with every oven reserved for them, and offers them a piece of the business. That is roughly the pull being described, and it explains why prestige and salary alone were not enough to keep them.
Why it matters: the AI race is often told through model launches and benchmark wars, but the deeper contest is for a few hundred people who can actually build the next generation. This week, the flow ran out of the company that arguably started modern AI and into the two labs that are now operating under government-vetted release programs, the same labs in today's news about GPT-5.6 and Mythos 5. Talent and frontier access are concentrating in the same two places at once.
The honest caveat: be skeptical of the financial drama. The widely repeated claim that the departures erased a quarter-trillion dollars from Alphabet's market value rests on a single rough day for the stock, and stocks move for many reasons at once. Pinning a one-day swing on two resignations makes a tidy headline but is almost certainly an oversimplification; markets were digesting plenty of other news that week. The departures are real and confirmed through multiple outlets. The precise dollar figure attached to them is the shakiest part of the story, and you should treat it as narrative color, not a measured fact.