DeepMind's entire AlphaFold team disbanded; Nobel laureate and staff joining Anthropic; resources redirected to Gemini

The legendary team that won the Nobel Prize and created AlphaFold has been dissolved in place at Google DeepMind.

According to the Financial Times, this team—which conquered the protein structure prediction challenge and produced a Nobel laureate in chemistry—no longer exists as an independent unit. Its original members have been redistributed across multiple projects within Google: some transitioned to Gemini and AI programming, others to research on genomics, protein and enzyme design, and nuclear fusion. Still others moved to Isomorphic Labs, Alphabet's drug discovery subsidiary, while some departed Google entirely.

Pushmeet Kohli, Chief Scientist at Google Cloud and DeepMind's Vice President of Science, explained that DeepMind has always encouraged scientists to independently choose the problems they want to tackle, and AlphaFold team members have the same freedom.

In essence, after completing AlphaFold, Google allowed team members to carry their accumulated expertise into additional projects through internal mobility. Part of the team remained within DeepMind, transitioning to work on Gemini, AI programming, genomics research, protein and enzyme design, and even nuclear fusion research.

Isomorphic Labs, founded by Demis Hassabis in 2021, was created with the goal of advancing the capabilities demonstrated by AlphaFold toward drug discovery and design. From this perspective, AlphaFold hasn't been shelved—it has simply evolved from a research project aimed at scientific breakthroughs and publications into work more directly connected to drug development.

In 2024, the AlphaFold creator just won the Nobel Prize in Chemistry for protein structure prediction work, sharing the honor with Hassabis.

Two other key AlphaFold researchers—Jonas Adler and Alexander Pritzel—are also moving to Anthropic. Their work extended beyond AlphaFold: Adler participated in Google's AI programming initiative, while Pritzel contributed to Gemini model training. However, both were listed among AlphaFold's core authors.

According to Financial Times tallies, nearly a quarter of the DeepMind employees who originally authored the AlphaFold paper have now left. But the Nobel laureate and multiple core researchers—at roughly the same time—flowed to the same competitor: Anthropic.

The AlphaFold team's dissolution generated such significant attention because what it accomplished carries real weight. The team solved a protein structure prediction problem that had troubled the biology field for decades, produced predictions for over 200 million protein structures, and gave rise to two Nobel laureates in chemistry. The Nobel Committee's assessment of AlphaFold 2 was that it can now predict the structure of almost all known proteins.

DeepMind assembled a small, tightly integrated cross-disciplinary team—machine learning researchers, biologists, and engineers—who focused exclusively on a single problem for an extended period. When the first generation of AlphaFold hit a plateau, rather than patching and refining the existing approach, they scrapped the old design and rebuilt from the model architecture up. This approach was like assembling a mountaineering expedition: the peak was in sight, everyone focused on the same summit, and the team pressed upward despite the challenge.

Based on where original team members have gone, DeepMind appears to be dismantling the AlphaFold team and distributing its accumulated talent and methodologies across additional scientific problems. Scientists no longer need to train a custom model from scratch for each new challenge. Gemini can first process literature, data, and code, then work alongside AI Co-Scientist agents and tools like AlphaEvolve to help researchers formulate hypotheses, refine algorithms, and filter solutions.

The newly launched Science Skills feature already connects over thirty life sciences databases and tools, including the AlphaFold database, AlphaGenome, and UniProt. A single AlphaFold team could tackle one problem at a time; a research platform built on Gemini can simultaneously serve biology, materials science, mathematics, nuclear fusion, and other disciplines. Researchers no longer need to rebuild modeling capabilities, compute infrastructure, and research tools for each new project, and more scientists can access analysis methods that previously had high barriers to entry.

However, this approach carries hidden risks. AlphaFold succeeded because a small group of people was willing to focus on what seemed like an unsolvable problem for years. That kind of concentration is difficult to platform. Once the operation expands, the question becomes: how much attention can each individual problem receive? Will Google even remain willing to support a team working several years without deliverable products, solely to solve a single difficult challenge? With DeepMind now having disbanded this small team, whether it can identify the next problem worth spending a decade to solve remains the true open question.