John Jumper, the computational chemist who shared the 2024 Nobel Prize in Chemistry for building AlphaFold, is leaving Google DeepMind to join rival lab Anthropic. He announced the decision on Friday, June 20, 2026, closing a run of nearly nine years at the lab where he turned protein structure prediction from a decades-old research problem into a solved one. His exit lands in the middle of a recruiting fight between the best-funded AI companies, and it follows two other high-profile departures from DeepMind within the same span of weeks.
- John Jumper, co-creator of AlphaFold and a 2024 chemistry Nobel laureate, is leaving Google DeepMind for Anthropic after nearly nine years.
- His move follows Gemini co-lead Noam Shazeer joining OpenAI and AlphaGo researcher David Silver leaving to found his own startup.
- AlphaFold has predicted the structures of more than 200 million proteins, reshaping how biology and drug discovery work.
- Neither Jumper nor Anthropic has detailed what he will work on, and he plans to take time off before starting.
The news was first reported by CNBC and confirmed by Jumper himself in a post on X, then picked up across the trade press including TechCrunch and The Decoder. What makes it stand out is not just the prestige attached to Jumper's name. It is the pattern. Three of the most recognizable researchers Google DeepMind employed have walked out the door in a matter of months, two of them straight into the arms of the labs Google is racing against.
Who John Jumper is and why his name carries weight
John Jumper is not a household name in the way that some AI executives have become, but inside the field he is about as decorated as a working scientist gets. He joined the Google DeepMind protein team after finishing his PhD, and as he noted in his farewell message, DeepMind chief executive Demis Hassabis handed him the lead on the AlphaFold project just six months later. That bet paid off in a way that almost no research wager ever does.
AlphaFold attacked a question biologists had wrestled with for roughly fifty years: given the genetic sequence that codes for a protein, can you predict the three-dimensional shape it folds into? Shape determines function in biology, so an accurate predictor is enormously useful for understanding disease, designing drugs, and studying basic cell mechanics. Experimental methods to determine a single structure could take months or years of lab work. AlphaFold collapsed that to minutes for many proteins, and it did so at a level of accuracy that surprised even skeptics when the system swept the CASP structure-prediction contest.
The scale of what followed is hard to overstate. AlphaFold has now predicted the structures of more than 200 million proteins, a catalog that covers almost every organism with a sequenced genome. Researchers around the world treat the resulting database as basic infrastructure, the way an earlier generation treated published genome sequences. In 2024 the Royal Swedish Academy of Sciences recognized that contribution, awarding the chemistry Nobel jointly to Jumper and Hassabis for the AlphaFold work, alongside David Baker, who was honored for separate protein-design research.
So when Jumper changes employers, it is not a routine engineer reshuffle. It is one of the most credentialed names in AI-for-science choosing to do that science somewhere else.
The work also reshaped how a generation of biologists spends its time. Before AlphaFold, a structural biology lab might orient a multi-year project around solving one protein structure through X-ray crystallography or cryo-electron microscopy. With a strong prediction available in advance, researchers can skip ahead to questions about how a protein binds and what a mutation does to it, using the experiment to confirm rather than to discover the basic shape. Drug discovery teams now fold the predictions into early screening to narrow candidate molecules faster. That change in daily workflow, more than any single headline, is why the protein database is treated as a public good rather than a commercial product.
What John Jumper said about the move
Jumper kept his announcement gracious. Writing on X, he said he had decided to leave Google DeepMind and join Anthropic after taking some time to recharge. He thanked the lab directly, recalling that Hassabis "took a real chance letting me lead the AlphaFold team just six months after finishing my PhD." He added that "GDM is a special place, and I'll still be excited to hear about what amazing things they discover next."
Hassabis responded publicly and warmly. He described the partnership behind AlphaFold as extraordinary and said the system had changed the world, a fair claim given how widely the protein database is used. The tone on both sides was the opposite of acrimonious, which matters for a field where reputations and future collaborations are currency. Departures handled this way leave the door open for joint work later, something AI-for-science researchers tend to value because the problems are bigger than any single lab.
One detail both Jumper and Anthropic have left vague is what exactly he will do once he arrives. The public statements name the destination without naming the project. Anthropic has built its reputation on language models and AI safety research rather than computational biology, so a researcher whose signature achievement is protein folding raises an obvious question about fit. It is possible Anthropic wants his methods expertise rather than his specific domain, or that the company is signaling a broader scientific ambition. For now that is speculation, and the firms have not filled in the blank.
John Jumper is the third major name to exit DeepMind
The reason this single hire reads as a trend is the company it keeps. Within the same short window, Google DeepMind lost two other researchers whose names appear on foundational AI work.
Days before Jumper's announcement, Noam Shazeer said he would leave to join OpenAI. Shazeer is a co-author of the 2017 paper "Attention Is All You Need," the work that introduced the Transformer architecture underneath virtually every large language model in use today. He had been serving as a vice president of engineering at Google and a co-lead on the Gemini models, including the reasoning techniques in Google's recent releases. His path back to Google was itself expensive. In 2024 Google paid a reported 2.7 billion dollars in a deal tied to Character.AI, the startup Shazeer co-founded after an earlier departure, an arrangement that brought him and collaborator Daniel De Freitas back into the DeepMind fold. Now he is leaving again, this time for the company preparing one of the most watched IPOs in tech.
Weeks earlier, David Silver left as well. Silver led the research behind AlphaGo and AlphaZero, the systems that beat the best human players at Go and then taught themselves chess and shogi from scratch. According to The Decoder, he departed to start his own company focused on world models and reinforcement learning, two areas many researchers believe are central to the next phase of AI capability. Unlike Jumper and Shazeer, Silver is not joining a rival. He is building one.
Stack those three together and the picture is stark. The architect of the Transformer, the lead behind game-playing reinforcement learning, and the Nobel-winning force behind protein folding all left the same organization inside a few months. Each represented a different pillar of DeepMind's research identity.
Why the AI talent contest has become so fierce
None of this happens in a vacuum. The competition for senior AI researchers has reached a level that looks less like ordinary hiring and more like a market for scarce, almost irreplaceable assets. There is a small population of people who have actually shipped a defining AI system, and every well-capitalized lab wants them.
The economics behind that scramble are real. Anthropic and OpenAI have both raised enormous sums and command valuations that would have seemed absurd a few years ago. That funding translates directly into the budgets they can offer, in cash, equity, and the promise of compute at a scale individual academics cannot dream of. For a researcher weighing a move, the pitch is not only money. It is access to the clusters and the engineering support needed to chase an ambitious agenda. When a lab can credibly say it will give you more resources to pursue the work you care about, prestige and stability at an incumbent start to matter less.
Jumper said he plans to take time to recharge before starting at Anthropic, so his first contributions there may not appear for some time. The announcement marks intent, not an immediate change in output.
Researcher mobility is not new to AI, but the stakes attached to each move have climbed. Shazeer's own history illustrates the pattern. He left Google in 2021 to co-found Character.AI, then returned in 2024 through a deal that valued his return at billions, and is now leaving once more for OpenAI. People at this level rarely stay put for an entire career, and the labs have learned to plan around that churn. What has changed is the price. A single hire can now carry a reported nine- or ten-figure cost when it involves acquihires or large equity grants, which turns each individual decision into a board-level event rather than a routine offer letter. When the most senior researchers treat a move as a multi-year bet on where the frontier will be built, the receiving lab gains not only their hands but a signal to the rest of the market about which direction is worth backing.
There is also a competitive-intelligence dimension that the labs rarely discuss openly. Hiring a senior researcher does more than add capacity. It removes that person from a rival and brings along a mental map of how the competitor approaches problems. In a field where the gap between the top systems is often measured in months, that kind of knowledge transfer has value beyond the individual's direct output. The poaching is partly about building and partly about denying.
What the departures mean for Google DeepMind
Google has not been quiet during this period, and it would be a mistake to read three exits as a collapse. DeepMind remains one of the deepest research benches in the world, with thousands of scientists and a long record of producing systems that move the field. AlphaFold itself continues, and the protein database does not stop being useful because its original lead moved on. Institutions of this size are built to survive the loss of even their most visible people.
Still, the timing is uncomfortable. The departures coincide with a stretch where Google is under pressure to show that its Gemini line can stay ahead of Anthropic and OpenAI on the capabilities that customers and developers care about. The Decoder reported that insiders worried the then-upcoming Gemini 3.5 Pro release, expected in late June 2026, might not clearly outmatch the latest offerings from its two main rivals. Whether or not that specific concern holds up, losing marquee researchers while fighting a perception battle about model quality is the kind of one-two that makes a hard moment harder.
Bloomberg, cited by TechCrunch, has also reported that Google has struggled to commercialize its coding tools, an area where Jumper was described as a contributor. Commercialization stumbles and talent flight are not the same problem, but they feed the same narrative: that even a lab with DeepMind's pedigree can find itself on the back foot when the competition is this well-funded and this aggressive about hiring.
Retention strategy matters here too. Google has historically leaned on a mix of compensation, research freedom, and the gravity of working alongside other elite scientists to keep its bench intact. The fact that those levers did not hold three of its most visible people suggests the pull from rivals has grown strong enough to overcome them, at least at the very top of the seniority ladder. Whether that pull reaches deeper into the organization is the question Google's leadership has to weigh, because the researchers one rung below the Nobel tier are exactly the people the next round of recruiting will target.
What Anthropic gains, and what it signals
For Anthropic, landing Jumper is a statement as much as a staffing decision. The company has positioned itself around large language models and a heavy emphasis on AI safety, releasing its Claude model family and publishing research on interpretability and alignment. A Nobel laureate from outside that core competency is an unusual addition, and it invites a question about where Anthropic sees itself going.
One reading is that Anthropic wants to broaden into AI for science, treating Jumper's arrival as a seed for work that goes beyond chatbots and coding assistants. Another is more about talent gravity than strategy: signing a researcher of Jumper's stature helps recruit the next tier of scientists, who notice where the most respected people in the field choose to work. Both readings can be true at once. What is not yet known is which projects he will touch, and until Anthropic says more, the strategic meaning stays partly open.
The contrast with David Silver's choice is worth holding in mind. Two of these departing researchers concluded that the best place to do their next work is inside a large, well-funded lab. The third decided the better path was to start something new. That split reflects a genuine debate in the field about whether breakthrough research now requires the compute and capital that only a handful of companies control, or whether a focused startup can still carve out room on a specific bet like world models. The answer will shape where the next generation of systems comes from.
A field defined by where its people go
The story of John Jumper leaving DeepMind for Anthropic is, on its surface, a single hire. Read against the backdrop of Shazeer heading to OpenAI and Silver striking out on his own, it becomes something larger: a snapshot of an industry where the most valuable resource is a few hundred people, and where the balance of power can shift based on which of them takes which meeting. The labs are not only competing on models and benchmarks. They are competing on who they can convince to walk through the door.
What happens next will say more than the announcements did. If Anthropic builds a serious AI-for-science effort around Jumper, the move will look like the start of a new direction. If he ends up applying his methods knowledge to the company's existing priorities, it will look more like a prestige acquisition with a long fuse. Either way, the underlying lesson is one the AI industry keeps relearning in 2026: capability lives in people, and people, even Nobel laureates, are choosing their next labs with the whole field watching. For more on how the major AI companies are positioning against one another, see our ongoing coverage on the MaShop blog.