Google DeepMind dismantles AlphaFold team as it pivots to Gemini

Google DeepMind has dismantled the dedicated AlphaFold research team and reassigned most members to broader projects, the company told the press and staff, marking a strategic pivot toward its Gemini frontier models and away from one-problem science units.

The restructuring, first detailed by multiple outlets, moves AlphaFold researchers into Gemini-led programmes, enzyme design, genomics and other Alphabet units while some staff have left the company, according to reporting. DeepMind says AlphaFold’s work will continue, but no longer as a standalone team built around protein folding.

Nearly a quarter of original AlphaFold authors have departed

Financial Times’ coverage — relayed in regional reporting — says almost one-quarter of the full-time authors on the original AlphaFold papers have left since the reorganisation began, and that “most” remaining researchers were reassigned to other projects. Secondary accounts add that a portion of those departures have landed at rivals and startups working on large models and scientific AI.

The change carries symbolic weight: AlphaFold’s breakthroughs in protein-structure prediction earned prizes and became a central piece of DeepMind’s scientific reputation since its 2018 development. Company communications, however, stress this is a structural reshuffle rather than a termination. “Our strategy over the last nine years has been to focus on grand challenges... The strategy has evolved,” Pushmeet Kohli, DeepMind’s vice president of research, told reporters in comments cited in coverage of the move.

Shift toward Gemini-centred science and internal drug units

DeepMind frames the reorganisation as a move from organising teams around single scientific problems to building broader, Gemini-based systems that can assist multiple domains. Reporting says researchers were redeployed into Gemini-led science work, enzyme design, nuclear fusion and genomics, while some moved into Alphabet’s drug-discovery arm, Isomorphic Labs.

That explanation aligns with Google’s corporate priority: heavyweight frontier models are now core to product and research strategy. But the trade-off is clear. AlphaFold was a high-visibility, demonstrably useful scientific product; folding its expertise into general systems risks diluting specialised momentum even as it promises cross-domain reuse.

Competitors and talent flight reshape the calculus

Industry observers pointed to competition from frontier-model labs as a proximate cause of the timing. DeepMind’s pivot places it in the same strategic lane as other firms concentrating firepower on general-purpose agents and large models rather than single-solution teams.

Sceptics flagged two dangers. First, consolidating AlphaFold expertise inside Gemini could slow incremental, domain-specific improvements that users and partners depend on. Second, the reported departures — including to other AI labs — intensify a talent drain at a moment when specialised scientific credibility still matters to collaborators in pharma and academia.

Some coverage uses the language “shut down” or “dismantled”; DeepMind insists AlphaFold research continues within broader programmes. Those two frames are not mutually exclusive: a team can be dismantled administratively while underlying work proceeds under different management and cross-cutting platforms.

The move also has a commercial dimension. Embedding AlphaFold capabilities into Gemini or into Alphabet’s drug units could feed productisation and revenue-generation paths more directly than a standalone research team. But it also transfers control from a celebrated scientific brand to a corporate platform whose success is measured by competing metrics.

Closing: what to watch next

Watch two near-term signals: whether DeepMind publishes follow-up AlphaFold-style papers under Gemini authorship, and the pace of feature updates to public AlphaFold resources. Those will reveal whether the shift preserves the team’s scientific output or simply rebrands it as a component of a broader frontier-model push.