On Wednesday, August 7, Google announced a major reorganization of its AI division. Demis Hassabis, who founded DeepMind in 2010 and led it through its $650 million acquisition by Google in 2014, is stepping back from day-to-day CEO duties to become DeepMind Chair and Alphabet Chief Scientist, citing the imminence of AGI as his motivation. Koray Kavukcuoglu, the lab's chief technology officer, takes over as Senior Vice President of Google DeepMind reporting to Sundar Pichai, becoming responsible for Gemini model development.
The shakeup extends beyond DeepMind. Jeff Dean, Google's 30th employee and de facto chief scientist, is departing after 27 years to found his own AI startup alongside Oriol Vinyals and other Gemini co-leads. Dean co-founded Google Brain in 2011 and helped lead transformer research before the 2023 merger of Brain and DeepMind into Google DeepMind. His departure, alongside Vinyals and others, signals fracture in AI leadership that had already suffered from geographic and organizational tensions between the London-based DeepMind and Mountain View's engineering culture.
The organizational shift consolidates AI leadership toward California. Kavukcuoglu relocated from London to Google's headquarters in 2025. Sebastian Borgeaud, who leads a critical coding effort, also moved from UK to California. Alphabet's stock fell 4–5.5% on the news, reflecting investor concerns that Google is losing momentum in the race against OpenAI and Anthropic. London-based DeepMind, once the UK's flagship AI asset, now reports to a US-based executive, signaling a broader recentralization of Google's AI strategy toward the US mainland.
Architects and infrastructure leaders should track the organizational aftermath. Jeff Dean and Discovery Loop (his newly formed venture) claim "true full-stack depth" across chips, hardware, software, and models—a direct echo of Google's own stated strengths. Whether DeepMind's scientific pipeline (AlphaFold, protein research, drug discovery via Isomorphic Labs) can sustain momentum under Hassabis-as-chair minus his operational grip remains an open question for the AI safety and research community.