Google shook up the leadership of its AI operation on Wednesday, moving DeepMind co-founder Demis Hassabis out of day-to-day management and losing chief scientist Jeff Dean after a 27-year run at the company.
The changes were laid out in a memo from Alphabet chief executive Sundar Pichai posted to the company blog. Pichai said Hassabis will become chairman of Google DeepMind and chief scientist of Alphabet, while Koray Kavukcuoglu, the unit's chief technology officer, steps up to run DeepMind day to day as senior vice president. Kavukcuoglu reports straight to Pichai and takes responsibility for Gemini model development plus frontier AI research.
In the same memo, Pichai confirmed that Jeff Dean, the engineer credited with some of Google's biggest technical breakthroughs, is leaving to start a new company. Dean will join senior fellow Sanjay Ghemawat to build a public benefit corporation called Discovery Loop, aimed at using machine learning to speed up discoveries in science and engineering. Alphabet will act as a founding investor and supply cloud computing capacity to the venture.
News of the upheaval knocked Alphabet shares down about 4% in a single session. The timing is awkward for a company trying to reassure Wall Street that its outsized bets on AI infrastructure are paying off in growth.
A changing of the guard
The reshuffle is the clearest sign yet of how differently Google now organizes its AI work compared with the early years of the boom. DeepMind once operated with a fair amount of independence from the rest of the search giant. That changed when Google merged its Brain team into the lab and folded the combined unit into its wider AI effort.
Kavukcuoglu has been at DeepMind since 2010, arriving before Google bought the lab in 2014. He helped start the deep learning team and led work on voice synthesis research such as WaveNet. In recent roles he oversaw Gemini development as the lab's technology chief and served as Alphabet's head of AI engineering, a position that let him connect research and product work across the company.
Under him, Gemini stays on the same track. The company is pointing toward Gemini 4 as its next major release, after shipping three new Gemini models in the past few days, including one aimed squarely at the cybersecurity market. Gemini 3.5 Pro, the flagship Google hoped to deliver by the middle of last year, remains delayed.
Why the exits matter
Dean's departure carries weight because he has been part of Google's identity for so long. He helped build the search infrastructure that turned the company into a giant, then became one of the strongest voices pushing it toward neural networks at a time when much of the industry still had to be convinced those methods would work. His work spans search rankings, speech recognition, and the decision to build the TensorFlow project that underpinned years of Google research.
His new venture, Discovery Loop, sits outside Google. Dean told The New York Times that operating away from a public company gives the team room to make calls that are not always in the purest financial interest of shareholders. He and Ghemawat structured the project around machine learning and scientific discovery rather than a consumer product or a workplace tool.
The exodus goes beyond the two co-founders. Google has watched a steady stream of senior people leave over the past year or two. Gemini co-lead Noam Shazeer left for OpenAI, and Nobel laureate John Jumper, whose AlphaFold work helped secure the 2024 chemistry award, moved to Anthropic. The departures have raised quiet questions inside the company about whether it can hold onto the researchers who keep it near the frontier.
Even so, Google still commands enormous resources. Its cloud unit is growing faster than either Amazon Web Services or Azure, and revenue there climbed 82% in the second quarter to $24.8 billion. That makes the demand story about AI clearer than a promise; the money shows up in the bills the cloud business already sends to customers.
What changes, and what does not
For most companies doing business with Google, the practical effect of the reshuffle is small. Even as the people running the models change titles, the research pipelines stay in place, the training clusters keep running, and the spending keeps climbing.
Pichai said on the most recent earnings call that Google is still supply strained on AI compute capacity, and that the company expects full-year capital spending of up to $205 billion. It reported negative cash flow for the first time in its history in the latest quarter, largely because of those same outlays.
That makes this leadership move even more interesting. The personnel decisions came alongside some of the same growth plans that leadership approved almost two years ago, so the priorities on the engineering side remain consistent with what the top engineers set out to build.
Holding it together
A close look at the departures shows how much of the company's machine-learning story was built by people who are now heading into new orbits. Dean worked on search rankings, on speech recognition, and on the decision to build the TensorFlow framework, and became the most visible internal advocate of that direction.
His exit, combined with the promotion of Kavukcuoglu, makes clear that Google wants the next generation of DeepMind leadership to come from people who stayed inside the organization rather than from founders who wanted to operate independently.
Hassabis told staff he would keep working with Pichai on strategic and global matters tied to the AGI, and would spend more time at Isomorphic Labs. He called AI drug discovery the top application of the technology for improving human health. Even from the chairman seat, he intends to keep shaping where Alphabet's many research dollars go next.
What is at stake
The bigger story here is not just one company, but how every major lab is deciding how far to separate daily operations from long-term research bets. Alphabet, a public company that must answer to investors, chose to promote its most visible scientist to a chairmanship rather than keep him in the trenchwork. Its competitors are making the same kinds of calls with their own leadership teams.
Google's answer to the skeptics is straightforward: its talent pool, compute capacity, and distribution are enough to keep it competitive with OpenAI and Anthropic even as those two firms keep rolling out newer models. That argument is now being tested with a new set of people in the control room and a top researcher setting up shop outside the building.
So the reshuffle matters for a few different reasons. It settles a question about who answers to whom in a business where the technology moves fast and the money moves faster. It tells us that Google wants the next chapter of DeepMind written by insiders rather than by founders who wanted their own startup, and it quietly confirms that the company is betting its future on the same cluster-first compute strategy even as its costs balloon.
Readers will also notice that the leadership churn is not unique to Google. OpenAI has gone through its own series of changes, Anthropic's executives have shifted roles, and several smaller labs have spun out of bigger ones in the past eighteen months. Taken together, the industry is normalizing around a simple model: a handful of wealthy backers, huge compute bills, and a small number of very bright people deciding which risks are worth taking next.
Reported by the Tech Desk. More coverage on artificial intelligence, cloud and edge computing, and semiconductors.