Google DeepMind Is Becoming Just Another Google Department, and Its Founder May Be on the Way Out

DeepMind, the London lab that gave the world AlphaGo and AlphaFold, appears to be losing the independence it has guarded for more than a decade. According to a Guardian report, a former Google manager says the “era of DeepMind as an independent actor is over,” and a current employee now describes the lab as “just another subdivision” of Google. Researcher Koray Kavukcuoglu takes over day-to-day operations from the United States, but without the CEO title. That detail says a lot: DeepMind no longer has a chief executive of its own.

Founder Demis Hassabis has been moved to chairman. Reporting from the newsletter Pathfounders says he wanted to leave at the same time as chief scientist Jeff Dean, and that Google feared an even sharper share price drop than the one that followed Dean’s exit. The chairman role, on this account, is a bridge to an orderly departure in the coming months. Hassabis is expected to spend more time on his scientific work and on Isomorphic Labs, the Alphabet drug discovery startup he founded. Meanwhile all Gemini development is consolidating in the Bay Area, and co-founder Sergey Brin is reportedly taking a firmer hand.

Here is where it gets interesting, because two serious analysts read the same facts in opposite directions. SemiAnalysis argues Google is quietly losing the race for the best models: years of cautious compute allocation and a risk-averse culture, they say, mean DeepMind is no longer a frontier lab. Their comparison is IBM retreating from personal computers into mainframe consulting. Tim O’Reilly makes the opposite case. He thinks Google is deliberately running a different race, betting on infrastructure rather than model leadership, and points to Edison losing to Westinghouse: the person who built the light bulb lost to the person who built the grid.

The numbers make both readings plausible. Gemini’s annualised revenue stood at $12 billion in the second quarter of 2026. Google Cloud, by comparison, is projected to bring in more than $73 billion in AI infrastructure revenue plus another $120 billion in TPU sales by the end of 2027. That is where the money currently sits. But the rumour mill points somewhere less flattering too: Gemini 3.1 Pro is still in preview, and 3.5 Pro, announced in May for a June release, has quietly been shelved. Both suggest real trouble training top-tier models, not just a strategic pivot.

What this means for you: In the short term, nothing. Gemini keeps working, the free tier keeps working, and Google’s efficient Flash models are arguably the best value in the market right now. What is worth watching over the next year is whether Google keeps shipping models that compete at the very top, or settles into being the company that sells everyone else the compute. Both are fine businesses. Only one of them keeps you at the front of the pack, and if you pick tools based on who is leading, that distinction will start to matter.

Sources

• The Guardian: Google’s Demis Hassabis shifts role as DeepMind changes
• The Decoder: Google dismantles Deepmind as Hassabis heads for the exit
• SemiAnalysis: Gemini is Cooked but GCP is Cooking
• Tim O’Reilly: Google’s Westinghouse Bet