Google DeepMind Disbands Nobel-Winning AlphaFold Team in Pivot to Gemini

In 2020, AlphaFold solved one of biology’s hardest standing problems—predicting the three-dimensional structure of proteins from amino acid sequences alone—and changed the course of biochemical research overnight. In 2024, the work earned Google DeepMind’s Demis Hassabis and team member John Jumper the Nobel Prize in Chemistry, the first time the award recognized an AI system’s scientific contribution. By July 2026, the team that built it had been dismantled.

A Financial Times report published July 29–30, 2026 confirmed that DeepMind has disbanded the dedicated AlphaFold research group. The restructuring unfolded gradually over the past year, largely below public radar, but the departure of its most prominent member made the shift impossible to ignore.

Jumper Goes to Anthropic

John Jumper, the vice president and engineering fellow who shared the Nobel with Hassabis, left Google DeepMind for Anthropic in June 2026. His co-authors on the landmark AlphaFold 2 paper—Jonas Adler and Alexander Pritzel—have also departed. The three formed the technical core of the AlphaFold effort and collectively represent the deepest concentration of protein-structure AI expertise that existed at any single organization.

Jumper’s arrival at Anthropic is the latest chapter in what has become one of the most consequential talent migrations in AI history. Over the past eighteen months, Anthropic has attracted a string of senior researchers from DeepMind: Nobel-level talent, leading robotics engineers, and key figures in interpretability and alignment research. The pattern is consistent enough that it is no longer incidental—it reflects a deliberate recruiting strategy and, presumably, an environment researchers find compelling enough to leave the Nobel-Prize-winning lab that trained them.

Google DeepMind declined to comment on the specific departures. In a statement to the Financial Times, it characterized the restructuring as a natural evolution, noting that AlphaFold’s scientific mission had largely been achieved.

What Happened to the Rest of the Team

Researchers who did not depart were redistributed across DeepMind’s existing programs. The destinations tell a story about DeepMind’s current strategic priorities: Gemini (the foundation model program), enzyme design, nuclear fusion (the Genie program), genomics, and Isomorphic Labs—the Alphabet spinout focused on AI-accelerated drug discovery that was seeded directly by AlphaFold’s protein-prediction capabilities.

The AlphaFold database remains fully operational. As of the disbanding, the database contains over 200 million predicted protein structures covering nearly every protein in the known proteome, freely accessible to academic researchers worldwide. The server continues to accept queries. In the most practical sense, AlphaFold’s scientific output has been permanently deployed—what changed is that DeepMind no longer has a team dedicated to extending it.

DeepMind’s Strategic Bet on Gemini

The disbanding reflects a broader organizational philosophy that has sharpened at DeepMind over the past two years: moving away from standalone teams dedicated to individual scientific “grand challenges”—AlphaFold, AlphaStar, WaveNet—toward a model in which Gemini serves as a general-purpose scientific intelligence platform, with specialized domain knowledge woven in rather than siloed in separate groups.

The logic is defensible. AlphaFold’s core insight—that protein structure prediction could be solved by a large transformer trained on sequence data—now looks like a special case of a more general principle: sufficiently large and capable foundation models can make progress on many scientific problems that previously required domain-specific architectures. If Gemini (or a derivative fine-tuned on biological data) can replicate and extend AlphaFold’s capabilities while also being useful for enzyme engineering, genomics, and drug-molecule generation, then a dedicated AlphaFold group is an inefficient use of talent.

The counterargument is that specialization still matters at the frontier. The problems AlphaFold 3 and hypothetical successors would have tackled—RNA structure prediction, protein complex assembly, dynamic folding under cellular conditions—are hard enough to require researchers who think about protein biochemistry every day. By distributing that expertise across multiple teams and programs, DeepMind risks diluting the concentration of knowledge that produced the original breakthrough.

The Talent War’s New Calculus

Jumper’s departure is the most visible data point in a longer trend. Google DeepMind remains one of the world’s best-resourced AI research organizations and continues to publish first-rate work. But the past eighteen months have demonstrated that competitive compensation and prestige are no longer sufficient to retain the most sought-after researchers when smaller, mission-driven labs like Anthropic can offer significant equity, deep alignment with safety-focused research cultures, and—increasingly—technical environments that researchers describe as more creatively free.

Anthropic now counts among its employees multiple former DeepMind principals, at least one Nobel laureate, and several people who have publicly described their reason for leaving as discomfort with the pace of deployment at large organizations. Whether that narrative accurately describes the environment at DeepMind is unknowable from the outside. What is knowable is that Anthropic has successfully converted it into a recruiting advantage that shows no sign of slowing.

For the broader scientific community that depends on AlphaFold, the immediate practical impact is limited: the database works, the API works, and EBI maintains the infrastructure. But the absence of a team focused on extending and improving the system means that future advances in protein AI will most likely come from academic groups, Isomorphic Labs, or new commercial entrants—not from the organization that won the Nobel Prize for inventing the field.

That outcome is itself a data point about how AI research is reorganizing: tools that once defined the frontier become infrastructure, and the researchers who built them move on to the next unsolved problem—whether inside the same organization, or not.