VECTOR WIREAI INTELLIGENCE
NVDA$1,847+3.2%MSFT$512+1.1%GOOGL$199-0.4%META$728+2.7%AMD$184-1.2%TSM$212+0.6%PLTR$98+4.1%AI IDX4,821+1.9%
PKT
SEEDRefresh Models Deals Regulatory Sources

Dean, Ghemawat, Vinyals, Le Exit Google to Found Discovery Loop; Hassabis Steps Back as DeepMind CEO

Jeff Dean, Sanjay Ghemawat, Oriol Vinyals, and Quoc Le leave Google to co-found Discovery Loop as Demis Hassabis steps back from DeepMind CEO role.

Vector Wire — AI-assisted editorial illustration

The simultaneous departure of Jeff Dean, Sanjay Ghemawat, Oriol Vinyals, and Quoc Le to co-found Discovery Loop, a public benefit corporation focused on automating scientific research, removes four of Google's most senior AI researchers in a single move1,14. Paired with Demis Hassabis stepping down as CEO of Google DeepMind to become chair and Alphabet's chief scientist, the reshuffle hands day-to-day control of DeepMind to Koray Kavukcuoglu, who will take over daily management of DeepMind and development of the next Gemini model2,12. ANALYSIS The combined moves concentrate operational authority closer to Google's product leadership while dispersing the research leaders who defined Google's AI identity into advisory or external roles.

Why it matters

Dean joined Google in 1999 as the company's 30th employee and served as chief scientist at Google DeepMind for 27 years3. Ghemawat, a Google senior fellow, worked alongside Dean for nearly three decades; together they created foundational distributed systems including MapReduce and Bigtable. Quoc Le was a founding member of Google Brain, and Oriol Vinyals served as a senior research scientist at Google DeepMind. The departure of this group does not remove a single executive — it removes an interconnected cluster of researchers whose collaborative history spans decades and whose joint work underpins core Google infrastructure.

The big picture

Google's commercial AI engine is running hot. Cloud revenue grew 82% in the second quarter. Pichai has said 90% of Fortune 100 companies are using Gemini Enterprise. The stock is up 16% this year after jumping 65% in 2025. By revenue metrics, Google's AI bet is paying off, yet the leadership changes are occurring at the same time — suggesting the tension is not about whether AI works commercially at Google but about what kind of AI Google prioritizes.

The Verge reported that the changes "likely signal deeper issues within Google's AI organization," including Hassabis' interest in longer-term research over shorter-term product shipping8. CNBC noted that the shift "highlights the tension between Hassabis' long-term scientific ambitions and Google's need to commercialize, monetize and scale its AI technology"6. Hassabis is moving away from Google DeepMind's day-to-day operations to focus on AGI strategy, scientific research, and Isomorphic Labs. The language from both Hassabis and outside observers points in the same direction: the operational tempo Google now demands from DeepMind is misaligned with the research-first culture its most senior scientists signed up for.

Kavukcuoglu, who takes over DeepMind operations, has emphasized execution. He said Google's Flash model delivers "frontier-level capabilities while running four times faster and more efficiently than comparable models". That framing — speed, efficiency, deployment — is the vocabulary of a product organization, not a pure-research lab, and it indicates the direction DeepMind operations will take under his leadership.

Between the lines

Discovery Loop's structure is revealing. Dean, Ghemawat, Vinyals, and Le incorporated it as a public benefit corporation whose mission is to automate machine learning, science, and engineering. The company said it plans to use algorithms to "initiate and iterate thousands of experiments simultaneously, with the goal of partially automating the research process and expanding the scale at which experimentation can be conducted". Discovery Loop is also interested in recursive self-improvement — using AI to help create more powerful AI. "While science and engineering have tremendously advanced society over past centuries, progress has traditionally relied on slow, sequential human iterations, creating a significant bottleneck," the company said in a press release. Google is investing in Discovery Loop. The amicable exit and Google's financial backing suggest the departure functions as an externalization of frontier research ambitions that no longer fit inside a product-driven DeepMind, rather than a hostile break.

Google has also delayed the launch of its newest flagship model, Gemini 3.5 Pro. Meanwhile, the company previously paid nearly $3 billion to bring back Noam Shazeer through an acquihire, and Shazeer subsequently left Google for OpenAI in June, less than two years after that deal. Google is spending billions to reacquire product-oriented talent while its most decorated research leaders leave voluntarily — a pattern consistent with a deliberate rebalancing of the organization's priorities.

Alphabet stock dropped after the announcement11. CNBC reported that the departures "add to questions about Google's ability to retain top talent and stay at the frontier". The Financial Times reported that Google is shifting AI power back to co-founder Sergey Brin as Hassabis steps aside4.

The talent drain extends beyond the Discovery Loop founders. John Jumper left DeepMind for Anthropic. Dean reportedly plans to serve as CEO of Discovery Loop.

What's next

Kavukcuoglu now faces a dual mandate: ship competitive products against OpenAI and Anthropic while retaining research credibility. Hassabis, as chair and chief scientist, retains strategic influence but has moved away from daily management. Discovery Loop, backed by Google's own capital, will test whether the automated-research thesis Dean and his co-founders articulated can produce results outside the infrastructure of a parent company whose stock market capitalization the sources place at $4 trillion. ANALYSIS The question facing Google is whether a product-first DeepMind can continue to produce the foundational research breakthroughs that made commercialization possible in the first place.

CORRECTIONS: none for this article · this piece updates automatically as the story develops · corrections policy & trail →