Google is dissolving Deepmind's independent structure, marking a significant strategic pivot for the company's AI research division. Demis Hassabis, Deepmind's founder and CEO, is expected to depart in the coming months as the lab loses operational autonomy.
Koray Kavukcuoglu will assume day-to-day leadership without the CEO title, signaling Google's shift toward tighter integration rather than independent research governance. All Gemini development moves to the Bay Area, consolidating operations under a single umbrella.
The restructuring reflects deeper problems. Google faces serious obstacles training frontier AI models, according to internal assessments. This gap persists despite Google's cloud business generating billions in revenue and providing infrastructure advantages competitors lack.
The strategic question now turns on Google's true intent. The company may deliberately be refocusing on infrastructure and computational capacity as the foundation for future breakthroughs. Alternatively, the reorganization could signal that Google cannot keep pace with frontier model leaders like OpenAI and Anthropic, despite advantages in compute and data.
Hassabis built Deepmind into a world-class research institution that achieved breakthrough results in protein folding and game-playing AI. His departure, combined with the loss of institutional independence, represents a major shift in how Google approaches AI development. The lab operated semi-autonomously within Google for years, maintaining its own research agenda and publication practices.
The timing matters. As competition in large language models intensifies, maintaining separate research cultures becomes harder to justify. Google's internal struggles with model training suggest the company needs better coordination between infrastructure teams and research groups, not more silos.
Whether this consolidation accelerates Google's AI capabilities or merely tidies organizational structure remains unclear. The company's track record shows mixed results. Google developed strong foundational models and deployed Gemini widely, yet OpenAI's ChatG