News

Google Reshuffles DeepMind Leadership as AI Race Shifts from Research to Execution

• By John K. Waters
• 08/06/2026

Google has overhauled the leadership of its artificial intelligence operations, moving Google DeepMind co-founder Demis Hassabis into a broader strategic role and placing longtime executive Koray Kavukcuoglu in charge of the company’s flagship models and AI products.

The reorganization comes as Google faces growing pressure to convert its research advantages and vast computing infrastructure into products that can compete with OpenAI and Anthropic. It also coincides with the departure of Jeff Dean, one of Google’s most influential engineers, and several other prominent researchers who are forming an independent AI company.

Hassabis will step away from his day-to-day responsibilities as chief executive of Google DeepMind to become chair of the unit and chief scientist of parent company Alphabet. He will continue to lead Isomorphic Labs, Alphabet’s AI drug-discovery business.

Kavukcuoglu, previously DeepMind’s chief technology officer and Google’s chief AI architect, will become senior vice president of Google DeepMind and report directly to Alphabet Chief Executive Sundar Pichai. His responsibilities will include Gemini model development, frontier AI research, and the teams responsible for the Gemini application and developer tools, Google said.

The structure puts one executive in charge of moving DeepMind’s research into products used by consumers, developers, and businesses. It also moves Hassabis away from the operational demands of running a large product organization and gives him more time to focus on artificial general intelligence, scientific research, and AI policy.

“I’ve decided that now is the right time for me to hand over my day-to-day operational responsibilities at GDM, so that I have the time and space to focus on the big picture,” Hassabis wrote in a message to employees published by Google.

The company described the transition as a way to advance its commercial AI work while allowing Hassabis to focus on the implications of increasingly capable systems. Hassabis said he believes artificial general intelligence is “close at hand,” although there is no industry consensus on how AGI should be defined or how close current systems are to achieving it.

From Research Lab to Google’s AI Engine
The reshuffle reflects how much Google DeepMind’s role has changed.

DeepMind began as an independent London research laboratory in 2010 and was acquired by Google in 2014. It established its reputation through research projects such as AlphaGo, which defeated a world champion in the game of Go, and AlphaFold, which predicts the three-dimensional structures of proteins.

Google combined DeepMind with its Google Brain research organization in 2023. The merger was intended to concentrate AI talent, computing infrastructure, and model development in a single unit. That combined organization subsequently became responsible for Gemini, the model family that now supports features across Search, Workspace, Cloud, and Google’s consumer AI applications. Google described the merger at the time as a way to develop more capable AI systems while improving safety and responsibility.

The latest changes take that integration further.

Kavukcuoglu has spent 13 years at DeepMind and founded its deep-learning team. He helped oversee research behind WaveNet, an early system for generating natural-sounding speech, and Deep Q-Networks, which combined deep learning with reinforcement learning. In his role as chief AI architect, he has also been responsible for integrating DeepMind technology into Google products. Google’s executive biography describes him as leading both model development and deployment at scale.

That combination of research and product experience appears central to the decision. Google does not lack AI discoveries, computing resources, or distribution. Its challenge has been coordinating those assets quickly enough to respond to smaller competitors that can make decisions without navigating a company of Google’s size.

Pichai told employees that Google needed to “continue to move fast and with clear purpose,” while acknowledging that there were areas in which the company needed to improve. He said Gemini Flash was seeing strong demand, while DeepMind’s Gemma models had surpassed 900 million downloads.

Google said the Gemini application now has more than 950 million monthly users. Company figures are not directly comparable with all rival usage measurements, but the number shows that Gemini has reached substantial distribution through Google’s products.

The unanswered question is whether a more centralized operating structure will make the DeepMind organization faster or add more pressure to prioritize near-term product releases over long-term research.

Jeff Dean Leaves After 27 Years
The leadership changes were accompanied by the departure of Jeff Dean, who has spent 27 years at Google and most recently served as chief scientist of Google DeepMind and Google Research.

Dean is leaving with Google Senior Fellow Sanjay Ghemawat and DeepMind researchers Quoc Le and Oriol Vinyals. The four are establishing Discovery Loop, a public benefit corporation that plans to use AI to automate parts of scientific and engineering research.

The company says it will initially focus on machine-learning research before applying its systems to other scientific disciplines. Its stated goal is to automate the cycle through which researchers propose experiments, implement them, evaluate the results, and use those findings to design the next experiment.

Google will remain involved as a founding investor and cloud provider. The two companies also plan to collaborate on research involving machine-learning systems and supporting infrastructure, Pichai said.

The arrangement allows Google to maintain a connection to the new venture, but Dean’s departure is still significant. He helped develop much of the distributed computing infrastructure that allowed Google to operate at internet scale and later co-founded Google Brain. His work spans systems such as MapReduce and Bigtable, as well as the neural-network infrastructure that helped support modern AI development.

Ghemawat has worked alongside Dean on many of those projects. Le was a founding member of Google Brain, while Vinyals has played a leading role in DeepMind’s model research. Together, the departures remove a concentration of institutional knowledge from Google’s AI operation.

Alphabet shares fell as much as 5.4% following reports of the changes, reflecting investor concern about the loss of senior technical talent and the continuing cost of the AI competition. The Wall Street Journal described the moves as a broad overhaul of Google’s AI leadership, while the Financial Times said the changes would bring DeepMind more closely into Google’s operations.

A Different Kind of AI Competition
The reorganization comes at a point when the frontier AI race is becoming less about isolated model demonstrations and more about execution.

The largest developers are competing over how quickly they can turn model improvements into coding tools, enterprise services, and consumer agents. They are also spending heavily on data centers and specialized processors, raising the financial cost of delays.

Google holds advantages that few competitors can match. It designs its own Tensor Processing Units, operates a global cloud platform, and controls products with billions of users. DeepMind gives it one of the industry’s most accomplished research organizations.

Those strengths can also create organizational complexity. Research teams may be rewarded for long-term breakthroughs, while product groups operate on shorter release schedules. Safety researchers can push for more testing as commercial teams push to ship. Integrating a frontier model into Search carries risks that a standalone chatbot company does not face.

The new leadership structure assigns Kavukcuoglu the task of managing those competing demands. Hassabis will remain influential, but his role will be more strategic and less operational. That could create clearer accountability for Gemini products while preserving Hassabis’ position as the public face of Google’s long-term AI ambitions.

It could also expose a divide between two versions of DeepMind.

One is the research laboratory that produced AlphaGo and AlphaFold. The other is what Hassabis called the “AI engine room of Google,” responsible for models and services distributed throughout the company.

Google is betting that it can preserve the first while making the second move faster. The departure of Dean and his colleagues makes that balance harder to maintain, even as their new company remains connected to Google.

The leadership changes do not indicate that Google is retreating from frontier AI. They show the opposite. Google is reorganizing one of its most important divisions around the belief that research leadership alone is no longer sufficient.

The next measure of DeepMind’s performance will not be another management announcement. It will be whether the new structure produces better models and gets them into Google’s products faster, without sacrificing the research culture that gave the company its advantage.