A Fundamental Reorganisation
Recent reports confirm that Google DeepMind is undertaking a significant restructuring of its research and development efforts. The most prominent example is the dissolution of the dedicated team behind AlphaFold, the celebrated AI model that won a Nobel
Prize for its work on protein structure prediction. This is not an isolated event but part of a broader pivot. For years, DeepMind's strategy, as articulated by research VP Pushmeet Kohli, was to focus on 'grand challenges'—assembling brilliant, dedicated teams to solve specific, monumental scientific problems like protein folding with AlphaFold or mastering the game of Go with AlphaGo. Now, that strategy has officially 'evolved'. Instead of maintaining siloed units for distinct problems, the lab is consolidating its immense talent and resources around building universal AI systems.
From Specialists to Generalists
The old approach was like having a hospital full of world-class specialists, each a leader in their specific field. The new strategy is to build a single, polymathic genius who can master all of those fields and more. The focus is now on large, general models, with the prime example being Gemini, Google's flagship AI. The idea is that instead of creating bespoke AI for individual tasks, the future lies in massive, multimodal systems that can be applied across a vast array of scientific and commercial domains. Researchers from the former AlphaFold team have been reassigned to projects based on Gemini, as well as other areas like nuclear fusion and genomics, demonstrating this new, integrated approach. The goal is no longer just to solve problems, but to build systems that help scientists solve their own problems, eventually automating parts of the research process itself.
The Race for General Intelligence
This strategic pivot is not happening in a vacuum. It is a direct response to the generative AI boom and intense competition from rivals like OpenAI and Anthropic. The entire industry is concluding that building ever-larger and more capable general models is the most direct path to Artificial General Intelligence (AGI)—AI that can perform cognitive tasks at or beyond human levels. DeepMind CEO Demis Hassabis has stated he believes AGI is likely just a few years away, creating a sense of urgency. This reorganisation is a high-stakes bet that consolidating resources to build a single, unified intelligence is the fastest way to win the AGI race and secure a competitive edge.
Risks and Departures
However, this bold new direction comes with significant risks and costs. One major consequence has been a 'brain drain' of top talent. Nearly a quarter of the original full-time DeepMind authors on the AlphaFold papers have left the company entirely. Most notably, Nobel laureate John Jumper, a core leader of the AlphaFold project, departed for rival lab Anthropic in June 2026, followed by other key researchers. These departures have sparked surprise internally and intensified scrutiny of DeepMind's ability to retain top talent. Critics of the new strategy worry that it may stifle the kind of focused, deep research that led to breakthroughs like AlphaFold in the first place, putting too many eggs in the single, large-model basket. The loss of key scientists to competitors like Anthropic, which is now launching its own science-focused AI tools, underscores the competitive risk of DeepMind's strategic shift.
