Google absorbs DeepMind as founder Demis Hassabis exits
Google is absorbing its DeepMind AI unit and preparing for the departure of founder Demis Hassabis, signaling a major shift from independent research toward centralized commercial development.

Google is restructuring its AI operations by integrating DeepMind directly into the parent company, effectively ending the lab's era of independence. Under the new structure, co-founder Demis Hassabis has transitioned to a chairman role to facilitate an orderly exit, allowing him to focus on scientific research and his pharmaceutical startup, Isomorphic Labs. Day-to-day operations are shifting to the United States under researcher Koray Kavukcuoglu, who will manage the division without the chief executive title. Meanwhile, Google co-founder Sergey Brin is taking a more active role, and all Gemini-related development is consolidating in the Bay Area.
This reorganization comes amid reports that Google is struggling to keep pace with rivals in frontier model development, particularly regarding coding performance. The release of Gemini 3.5 Pro, originally announced in May for a June launch, has been largely shelved, while Gemini 3.1 Pro remains restricted to a preview mode. Despite these development hurdles, Google is seeing massive financial returns from its broader AI ecosystem. While Gemini generated an annualized revenue of $12 billion in the second quarter of 2026, Google Cloud is projected to bring in more than $73 billion in AI infrastructure revenue alongside another $120 billion in TPU sales by the end of 2027.
For AI practitioners and developers, this restructuring signals a definitive pivot in Google's strategy. Instead of prioritizing raw frontier model capabilities, the company appears to be focusing heavily on infrastructure efficiency and commercialization, as seen with its highly optimized Flash models. Developers relying on Google's ecosystem may need to adjust to a landscape where cutting-edge, general-purpose frontier models take a backseat to highly scalable, cost-effective cloud infrastructure and specialized enterprise tools.
This is our own summary of reporting by The Decoder



