Google DeepMind could lose its independence, and founder Demis Hassabis could leave the company in the coming months, according to a report from The Guardian and information from Pathfounders. The restructuring concentrates AI development in the Bay Area and transfers operational leadership to Koray Kavukcuoglu, who takes over the lab's day-to-day operations from the US without the CEO title.
A former Google manager told The Guardian that the "era of DeepMind as an independent actor is over." A current employee at the lab said it is "just another subdivision" of the tech giant. Hassabis's departure, according to Pathfounders, would have taken place at the same time as Jeff Dean's, but Google feared an even bigger drop in its stock price. Hassabis was then "promoted" to chairman to enable the transition, and is expected to devote more time to scientific research and to the pharmaceutical startup Isomorphic Labs.
Restructuring divides experts
The restructuring raises two interpretations. The first, from consultancy SemiAnalysis, holds that Google is losing the race for the most advanced AI models. The lab is no longer a frontier reference, and the departures reflect years of cautious allocation of computing power and a bureaucratic, risk-averse culture. The company reportedly had advanced AI systems before ChatGPT, but did not put them in users' hands quickly.
The second reading, from Web 2.0 inventor Tim O'Reilly, suggests Google is competing in another race, focused on infrastructure rather than model leadership. O'Reilly compares the situation to the contest between Thomas Edison and George Westinghouse: Edison created the light bulb and the first power plant, but Westinghouse won by building an alternating-current electrical grid that brought electricity to all of society. Google's Flash models, optimized for efficiency, would fit this reading.
Financial indicators and delays
Gemini's annualized revenue reached US$ 12 billion in the second quarter of 2026. Google Cloud, for its part, is expected to generate more than US$ 73 billion in AI infrastructure revenue by the end of 2027, in addition to US$ 120 billion in TPU sales.
Despite the cloud sector's optimism, Google faces difficulties in developing frontier models. Gemini 3.1 Pro remains only in preview mode, and Gemini 3.5 Pro, announced in May for a June launch, has been essentially shelved. The problems point to failures in the previous team, which failed to catch up with the leaders, especially in coding performance. With the restructuring, Google intends to pursue leadership in both infrastructure and AI simultaneously, without preserving old structures.
