Google’s DeepMind is having an identity crisis

Google DeepMind has spent the year looking less like a steady research machine and more like a company in the middle of a rearrangement. Last week, cofounder Demis Hassabis stepped out of the CEO seat and into a chair role. Koray Kavukcuoglu, the company’s chief technology officer, took over the day-to-day job and now reports directly to Sundar Pichai instead of holding a separate DeepMind chief title.

Then Jeff Dean, DeepMind’s chief scientist, left just minutes later. That one-two punch made the changes feel bigger than a simple management shuffle. Inside Google, some engineers see the move as another step in a slow pull of power away from London, where DeepMind started, and toward Mountain View, where Google is based. Google pushes back on that reading, saying London still matters and that DeepMind keeps its research independence.

The personnel churn has not stopped there. In June, Google lost Gemini co-lead Noam Shazeer to OpenAI and AlphaFold co-inventor John Jumper to Anthropic. Two other AlphaFold veterans, Jonas Adler and Alexander Pritzel, followed him out. Engineers point to rival companies throwing money at talent, frustration over coding benchmarks, and the lure of pre-IPO equity before those rivals go public.

The model side has not made the mood easier. Gemini 3.5 Pro, which Google showed off at I/O in May, has missed three release dates. And according to Artificial Analysis, Google’s best shipped model, Gemini 3.6 Flash, is now behind Anthropic, OpenAI, xAI, Meta, and at least one Chinese lab on raw intelligence. That is a sharp reversal from last year, when Google briefly sat on top of the AI leaderboard.

Still, Google is not suddenly helpless. It has the chips, the cloud, and the distribution that AI-native rivals cannot match. But four years after Google’s first AI code red, DeepMind is again trying to prove it can move fast enough. This time, it has to do it without the two people most closely tied to its scientific identity.