Hassabis Steps Back as Google Restructures DeepMind Around Chip Politics

The Core · TL;DR
- Demis Hassabis is stepping back from day-to-day leadership of Google DeepMind and plans to leave entirely within months, becoming chairman while Koray Kavukcuoglu runs operations.
- DeepMind researchers report chronic TPU shortages even as Google Cloud sells the same chips to rivals like Anthropic, a dynamic researchers call a conflict of interest.
- Google Cloud's AI infrastructure business is projected to generate over $73 billion in revenue, with TPU sales alone reaching $120 billion by 2027.
- The AlphaFold research team has been broken up, with nearly a quarter of its members leaving Google DeepMind, as Gemini development consolidates in the Bay Area.
Demis Hassabis is leaving the operational helm of Google DeepMind, and the reasons researchers are giving reporters have less to do with ambition than with hardware. Internally, engineers describe a persistent shortage of access to TPUs, the very chips Google builds and sells.
That shortage sits at the center of a broader reshuffle. Google Brain and DeepMind were folded into a single "Google DeepMind" organization some time ago, and roughly a year before this reporting, Hassabis quietly handed day-to-day control to Koray Kavukcuoglu. Hassabis has now been elevated to chairman, a title change that appears designed to smooth an eventual full exit rather than signal expanded authority.
The chip complaint carries an unusual edge: Google Cloud sells TPU capacity to outside labs, including rival Anthropic, even as its own DeepMind researchers say they struggle to secure enough compute for their work. Google allocates infrastructure years ahead of need, but researchers note that priorities inside the company can shift abruptly, leaving teams scrambling.
Researchers describe the arrangement as a structural conflict of interest, with Google's cloud business effectively competing with its own research arm for the same scarce chips.
The commercial logic for prioritizing external sales is not hard to find. Google Cloud's AI infrastructure business is projected to top $73 billion in revenue, with TPU sales alone expected to reach $120 billion by the end of 2027. A newly announced deal with a company called Mirendil, worth more than $100 million in TPUs and Nvidia GPUs delivered through Google Cloud, underscores how central external chip sales have become to Google's business model.
Meanwhile Gemini, the product that increasingly defines Google's AI identity, is already generating an annualized $12 billion in revenue as of the second quarter of 2026. All Gemini development is now consolidating in the Bay Area, a geographic centralization that coincides with Hassabis's reduced footprint.
The restructuring has visible costs on the research side. The team behind AlphaFold, one of DeepMind's most celebrated scientific achievements, has been broken apart, with most of its researchers reassigned and nearly a quarter having left the company altogether.
Hassabis intends to depart DeepMind entirely within the coming months, redirecting his attention to his own scientific interests and to Isomorphic Labs, the drug-discovery venture he also leads. With his exit, Sergey Brin is expected to take on a larger direct role in shaping Google's AI strategy, a shift that places commercial and infrastructure priorities even more firmly in the hands of Google's core leadership rather than the research organization that built much of its AI reputation.
Original reporting and research used to synthesize this article.
WAKIB Editorial Team
This review was prepared and summarized by the WAKIB AI intelligence engine and vetted by our editorial board for accuracy and reliability.
Subscribe to Newsletter
Get a weekly summary of the most promising AI research and tools delivered to your inbox.
Telegram Channel
Join our active community on Telegram for real-time tracking of AI models and trends.
