Hook
The AlphaFold program that earned Google DeepMind a Nobel Prize in Chemistry is being wound down. According to the Financial Times, Google has reassigned most of the program’s key members and original paper authors to other projects, while a handful of senior researchers have already left the company. The move signals a quiet but consequential rebalancing of DeepMind’s research priorities, with the protein-folding work that defined a decade of computational biology now overshadowed by the company’s push into Gemini.
The team behind the breakthrough
AlphaFold was never just a model. It was a coordinated program that began inside DeepMind in 2018 and, in 2020, was recognized as a solution to the half-century-old “protein folding problem.” Scientists had spent decades working out the three-dimensional shapes of roughly 170,000 proteins using X-ray crystallography and nuclear magnetic resonance. AlphaFold took the data from that long arc of human work and used it to train an AI that could predict those structures in minutes.
In 2021, Nature published the methodology and the structure predictions of the entire human proteome. DeepMind then opened the AlphaFold Protein Structure Database, giving researchers free access to over 200 million protein structure predictions. The program became foundational infrastructure for drug discovery, vaccine development, and the study of neurodegenerative diseases such as Alzheimer’s and Parkinson’s.
The Nobel and the splintering
In 2024, the Royal Swedish Academy of Sciences awarded the Nobel Prize in Chemistry to DeepMind CEO Demis Hassabis and John Jumper, who had been a staff research scientist at the start of the AlphaFold program and rose to become a vice president and engineering fellow. The award cemented the program as a flagship of scientific AI.
Less than two years later, that team is no longer intact. Jumper announced in June 2026 that he was leaving DeepMind to join Anthropic, and several of his colleagues followed. DeepMind has confirmed to the Financial Times that it also moved staff members internally to Gemini-focused projects, while other former AlphaFold members were reassigned to Isomorphic Labs, the Alphabet-owned drug-discovery company that was spun out of DeepMind in 2021.
Why DeepMind is pivoting
Pushmeet Kohli, the vice president of research at Google DeepMind, told the Times that the strategic emphasis has shifted. “Our strategy over the last nine years has been to focus on grand challenges … a concrete goal every project is focused on,” he said. “The strategy has evolved.” The pivot effectively places Gemini, the company’s flagship large language model family, at the center of DeepMind’s research investments.
The decision lands against a backdrop of acute commercial pressure. Frontier AI labs are locked in a multi-front race to ship models that can match or exceed the latest releases from OpenAI, Anthropic, and others. Talent is the limiting reagent, and senior researchers across DeepMind have been courted aggressively. Internal reassignments to Gemini mean that the protein-folding work at the heart of AlphaFold is being shepherded rather than expanded.
The technical legacy that AlphaFold leaves behind
AlphaFold’s contribution to machine learning extended well beyond biology. Its Evoformer architecture, the attention-based module that processes both multiple sequence alignments and pairwise residue relationships, became a reference design for graph-structured prediction tasks. Subsequent iterations, including AlphaFold-Multimer for protein-protein complexes and AlphaFold’s updates for nucleic acid and small-molecule binding, have been integrated into commercial drug discovery pipelines at companies including Insilico Medicine, Isomorphic Labs, and Recursion.
The open release of the AlphaFold Protein Structure Database under a permissive license has been a critical factor in this adoption. Independent researchers have used it to accelerate vaccine design, identify new antibiotic candidates, and probe the structural basis of disease. That public resource is unaffected by the team restructuring. What changes is the rate at which the underlying models will be updated, and the agenda of the next research wave.
What drug discovery pipelines now look like
Isomorphic Labs, the Alphabet-owned drug discovery company that absorbed part of the AlphaFold team, is the most direct continuation of the project’s commercial ambitions. The company has used AlphaFold-derived models to partner with Novartis and Eli Lilly on small-molecule drug discovery, and the program has produced a portfolio of preclinical candidates targeting oncology and immune-mediated diseases. The pipeline work is one of the few places where AlphaFold’s research velocity is likely to continue at scale, even as the broader DeepMind program is wound down.
For independent researchers, the open AlphaFold database remains the single most important resource. The team changes do not affect the database itself, but they do indicate that future AlphaFold model releases will be slower, and that the next wave of protein structure prediction research will come from open-source groups and from academic labs building on top of the published AlphaFold architectures.
What the restructuring means for science
Scientific users of AlphaFold will not lose access to the protein structure database, which remains online for now. But the program’s future is no longer one of continuous expansion. The team that authored the original AlphaFold papers has been scattered across Anthropic, Isomorphic Labs, and internal Gemini projects. New releases of the model itself, which had been quietly upgraded several times since 2021, are likely to slow.
For the broader AI-for-science field, the change is a marker of how hard it is to sustain long-horizon research inside a frontier lab whose economics depend on quarterly model cycles. The same pressures that have pushed Alphabet to pour tens of billions of dollars into Gemini infrastructure are now pulling the most senior protein-folding experts away from the work that brought them a Nobel Prize.
Conclusion
The winding down of AlphaFold is not the end of computational structural biology. Open-source alternatives and academic consortia have begun to fill the gap, and the protein structure database remains a resource for the field. What changes is the gravitational center inside DeepMind itself. The unit that won a Nobel in 2024 is being absorbed, in pieces, into a Gemini-first research agenda. The protein-folding problem is solved, in the practical sense that AlphaFold solved it. What comes next for the team that solved it is a different kind of question, and the answer, for now, lives at Anthropic, at Isomorphic Labs, and inside Gemini.