Less than a year ago, AlphaFold won Google DeepMind a Nobel Prize. Now DeepMind has taken the team that built it apart.
The lab has reassigned most of the original AlphaFold-paper authors over the past year, the Financial Times reported. Nearly a quarter of them have left the company altogether. DeepMind confirmed the moves, and is folding the work into a wider push around Gemini.
From one grand challenge to the ‘AI scientist’
AlphaFold was the purest expression of DeepMind’s old strategy: point a dedicated team at one hard problem and solve it. It cracked the 50-year protein-folding problem, Engadget noted, predicting 3D protein shapes in minutes. It now underpins drug discovery and disease research worldwide.
That model is now on the way out. “Our strategy over the last nine years has been to focus on grand challenges,” DeepMind research VP Pushmeet Kohli told the FT. “The strategy has evolved.” Instead of one team per problem, the lab is now building Gemini-powered systems meant to assist scientists and automate parts of research itself.
Former AlphaFold staff have scattered. Some moved to Gemini projects, others to enzyme design, nuclear fusion and genomics, The Decoder reported. A few went to Isomorphic Labs, the Alphabet drug-discovery spinout that AlphaFold inspired.
The stars went to Anthropic
The most painful losses walked out entirely. John Jumper, who shared the 2024 Nobel with DeepMind chief Demis Hassabis, left for Anthropic in June. Two core AlphaFold researchers, Jonas Adler and Alexander Pritzel, followed him.
One DeepMind employee told the FT the three were “instrumental, important, core members” whose exits “sparked surprise internally.” They are now at a rival that just launched Claude Science, a workbench built for the exact biology and drug-discovery work AlphaFold pioneered.
Why it matters
The kindest reading is that AlphaFold simply did its job. It solved its problem, spun off a company, and freed its people for the next thing. DeepMind says it remains “incredibly proud” of the work.
The harder reading is what the shift signals. The lab most defined by deep science is reorganising around the same language-model race as everyone else. Worse, it is losing the talent that made its name to the rivals driving that race. Betting on an “AI scientist” is a bigger, vaguer goal than folding a protein. Whether it pays off like AlphaFold did is an open question.
