Google has reassigned the key members of its AlphaFold team — the group behind one of the most important AI breakthroughs in biology — to other projects within the company, while some have left entirely. The move effectively dismantles the unit responsible for solving a problem that had stymied scientists for half a century. Engadget

The dismantling was first reported as an exclusive by the Financial Times, where it led the publication's Technology section. Financial Times

AlphaFold is an AI program that predicts the three-dimensional shape of a protein from its amino acid sequence — the chain of building blocks that defines each protein. Determining a protein's structure traditionally required years of lab work; AlphaFold can produce highly accurate predictions in minutes. Development began at DeepMind, Google's AI research arm, in 2018. By 2020, the scientific community recognized it as a solution to the 50-year-old protein folding problem. In 2021, Nature published both the methodology and structure predictions for the entire human proteome — the full set of proteins produced by human cells. DeepMind subsequently launched the AlphaFold Protein Structure Database, giving researchers free access to over 200 million protein structure predictions. Engadget

The recognition was capped in 2024, when DeepMind CEO Demis Hassabis and John Jumper received the Nobel Prize in Chemistry for their work on the system. Engadget

Jumper announced in June that he was leaving DeepMind to join Anthropic, a rival AI lab, and some of his AlphaFold colleagues made the same move. Other former team members were reassigned to Isomorphic Labs, the Alphabet drug-discovery spin-off from DeepMind. DeepMind confirmed to the Financial Times that remaining staff were moved internally to Gemini-focused projects. Engadget

Pushmeet Kohli, VP of Research at Google DeepMind, said the strategy has evolved from focusing on grand challenges. Engadget AlphaFold was a flagship example of DeepMind's grand-challenge approach — ambitious scientific problems tackled for their own sake. The implication is that it no longer fits the company's current priorities in the way it once did. DeepMind also unveiled AlphaFold 3, an AI model extending structure prediction to DNA, RNA, and ligands (small molecules that bind to proteins) in addition to proteins, though the fate of that lineage under the reorganization is unclear. Financial Times

Hassabis has stated that AlphaFold predictions were a key part of solving the overall protein structure in three out of four papers, underscoring the system's deep integration into mainstream structural biology research. Financial Times The Financial Times also published a podcast episode titled "AI Labs — Google DeepMind plans its comeback," discussing AlphaFold and the company's future direction. Financial Times

The reallocation follows a recognizable pattern: a large technology company consolidating talent around its highest-priority commercial bet. Gemini is Google's competitive front line against OpenAI and Anthropic in the foundation model race — the effort to build the most capable general-purpose AI systems. Pulling research staff from a solved scientific problem into that effort is a straightforward resource decision. The AlphaFold database remains publicly available, and the methodology is published. The research output persists even as the team behind it disperses.

What does not persist is the institutional concentration that produced iterative breakthroughs like AlphaFold 3. Isomorphic Labs absorbs part of that capability into a commercial drug-discovery pipeline, and Anthropic gains researchers with deep experience in applying transformer architectures — the neural network design underlying modern AI systems — to scientific domains. But the specific organizational unit that cracked protein folding no longer exists in any coherent form.

The broader context here is Kohli's framing about the shift away from grand challenges, and that phrase carries real weight. DeepMind's identity was built on grand-challenge demonstrations, from AlphaGo to AlphaFold, that served as both scientific milestones and recruitment magnets. If the company is genuinely de-emphasizing that model in favor of product-aligned research for Gemini, it signals a structural change in how one of the world's premier AI labs allocates its deepest technical talent. Whether that tradeoff pays off depends entirely on how competitive Gemini becomes, and how much of the scientific research momentum AlphaFold generated can continue without a dedicated team advancing it.

The AlphaFold database and published methods are durable artifacts. The team that could produce the next AlphaFold is now scattered across Gemini, Isomorphic Labs, and Anthropic.