Hassabis Leaves DeepMind CEO Role After Quiet Year-Long Handoff

Demis Hassabis formally left the Google DeepMind CEO job on Wednesday after at least a year of quietly handing day-to-day Gemini and consumer AI work to Koray Kavukcuoglu. He becomes Chair of Google DeepMind and Chief Scientist of Alphabet while he continues as founder and CEO of Isomorphic Labs.

Two people familiar with his thinking told Semafor the shift was not a forced exit. Hassabis had grown dissatisfied with pure tech-executive duties and more animated by scientific puzzles and AI safety.

The Roles That Changed Hands Wednesday

Alphabet CEO Sundar Pichai laid out the new map in Sundar Pichai’s note to DeepMind teams. Hassabis keeps close advisory ties to the models and research. Kavukcuoglu, already Chief AI Architect and CTO of Google DeepMind, steps up as SVP of Google DeepMind and reports directly to Pichai.

Kavukcuoglu now owns Gemini model development, frontier AI research, and the Gemini app plus developer teams. Hassabis will advise across those areas from DeepMind’s new London Platform 37 offices while he leans harder into Isomorphic and long-horizon AGI strategy.

• Gemini app: more than 950 million monthly users
• Gemma models: more than 900 million downloads
• Q2 revenue: up 24 percent to about $120 billion, cloud up 82 percent

Pichai also confirmed that longtime chief scientist Jeff Dean is leaving after 27 years with Senior Fellow Sanjay Ghemawat to launch an independent public benefit corporation. Google will be a founding investor and Cloud partner.

The Wednesday announcement therefore closed three open loops at once: the CEO seat, the chief-scientist seat, and the informal Gemini operating load that had already drifted to Kavukcuoglu. Each loop now has a named owner and a clearer reporting path.

Why the Scientist Quit the Executive Chair

Hassabis founded DeepMind in 2010 and sold it to Google in 2014. He shared the 2024 Nobel Prize in Chemistry for AlphaFold protein-structure work. Inside the company he had become the public face of AI, yet the people who know his thinking say the daily grind of product racing left him cold.

He told staff the moment felt decisive.

I’ve been working towards AGI my whole life and now, like many of you, I feel it is close at hand. It’s critical that we collectively get the next steps right to ensure this all goes well for humanity.

That line sat in his internal note and echoed in his own post announcing the Chair role, which drew more than 16,000 likes within hours. He added that he wants time and space for the big picture and to accelerate Isomorphic’s push against diseases such as cancer. One person who spoke with Semafor predicted he will be happier and may win another Nobel.

Pichai framed the same choice as letting Hassabis put full attention on “actively shaping the future of AGI,” work he called vital to Alphabet and humanity. The pair had discussed the new arrangement for a long time.

The Nobel work and the AGI timeline now sit on the same side of his calendar. Product surface decisions, once part of the CEO brief, move off that calendar so the scientific and safety questions can dominate it. That is the mechanism the people closest to him describe: less racing, more framing.

Koray Kavukcuoglu Now Holds the Product Keys

Kavukcuoglu joined DeepMind in its early days and has spent 13 years there. He started the deep-learning team and drove breakthroughs including WaveNet and DQN. He studied under Yann LeCun and held prior research roles at NEC Labs.

Koray Kavukcuoglu’s long DeepMind record made him the internal choice to keep Gemini delivery moving. He already carried much of the model and consumer load during the past year. Staff and leadership expect the formal title to remove friction rather than add it.

• Started DeepMind’s deep-learning team
• Key contributions to DQN (deep Q-network) and WaveNet audio generation
• Prior CTO of Google DeepMind and Google Chief AI Architect
• Now SVP overseeing Gemini models, frontier research, app and developer teams

In his own X reply he called the moment pivotal and said he was honored to lead while remaining tightly linked to Hassabis. Gemini 4 is the near-term test of whether the cleaner reporting line speeds shipping.

The cleaner line matters because the last year already showed the operating pattern. Kavukcuoglu absorbed the load; the SVP title and direct report to Pichai simply make that pattern official. Friction that once required a dual check with the CEO office can now clear on a single path.

Oriol Vinyals (DeepMind VP of research and Gemini technical lead) and Quoc Le (Google Brain cofounder, AutoML work) join Dean and Ghemawat as cofounders of the new lab.

Dean’s Exit Opens a Parallel Science Lab

Dean’s departure landed with more personal sadness inside Google than Hassabis’s role change. He had few direct reports and no large division, yet he shaped core infrastructure for more than two decades: MapReduce, Bigtable, TensorFlow, TPUs, Google Brain, and later Gemini technical leadership.

1. July 25, 2026, Dean speaks at Y Combinator Startup School about automated scientific-method loops across biology, chips and AI itself, without revealing his own plan.
2. Late July, The four cofounders organize Discovery Loop in stealth, meet VCs including Khosla Ventures and Radical Ventures on weekends and at Dean’s house.
3. August 5, 2026, Google announces the exits and its founding-investor plus Cloud-partner role; Pichai publicly thanks Dean and Ghemawat after trying to keep them.

Discovery Loop is structured as a public benefit corporation. Its first customer is itself: autonomous experiment loops aimed at improving machine-learning algorithms, then generalized to drug discovery, materials, chip design and other domains. Dean told Wired the idea crystallized only weeks earlier. The group wanted the freedom of a startup after years of large-organization inertia. Google still supplies compute for the first year and keeps a research collaboration path open.

The timing creates a second-order effect few market notes captured cleanly. Google now has an internal science-and-AGI chair in Hassabis, a product-and-model operator in Kavukcuoglu, and an external, Google-backed AI-for-science lab led by its former chief scientist. Talent that might have left for rivals can still sit inside Alphabet’s orbit.

How Markets and the AI Race Read the Day

Alphabet shares fell roughly 4 percent Wednesday, trading near $361 at one point, even though the stock remained up about 16 percent year to date. Investors already fretted about Gemini delays and record capital spending that pushed free cash flow negative for the first time since the 2004 IPO. Semafor noted Google’s models sit roughly six months behind the frontier on coding, the workload now consuming the most compute.

An internal executive told Semafor the management change should accelerate AI development rather than slow it. Hassabis had become the face of the effort yet was not focused on the product surface that sets the company’s standing in the consumer and developer race. Earlier exits, including Nobel co-winner John Jumper’s move to Anthropic, already raised recruitment questions. Structural shifts often produce more departures; whether that happens here remains open.

On X the loudest official voices were the principals themselves. Crowd reaction mixed respect for the science pivot with unease that two of Google’s most important AI figures were leaving day-to-day control on the same afternoon. Some observers treated the Chair title as a soft landing; others saw a deliberate split between product velocity and long-horizon AGI work. A few joked that founders finally stepped back so the operators could run.

The market move and the coding lag form the same pressure test. If the SVP line shortens decision cycles on Gemini, the lag narrative can ease. If senior research exits continue, the lag narrative hardens. Wednesday priced the uncertainty, not the design.

What the Split Changes Inside Google

The practical change is already a year old. Kavukcuoglu had been absorbing Gemini and consumer AI responsibilities. Formalizing the SVP role and direct Pichai line simply matches the org chart to reality. Hassabis keeps model and research advisory rights and still sits inside the full Alphabet AI stack: custom TPUs, Cloud, Search distribution, Android, and the Gemini app’s massive installed base.

Google’s own framing is momentum, not retreat. Flash demand is high, a Cyber model is live, and upcoming releases are described as close. Hassabis told staff the company remains the only one with a full stack world-class at every layer from infrastructure to frontier models to applications. He wants AI to prove “unequivocal value” first in human health.

What we know

• Hassabis is Chair of GDM and Alphabet Chief Scientist; he keeps Isomorphic Labs.
• Kavukcuoglu is SVP of GDM reporting to Pichai with Gemini and research ownership.
• Dean, Ghemawat, Vinyals and Quoc Le form Discovery Loop; Google invests and supplies Cloud.
• The handoff of day-to-day Gemini work began roughly a year earlier.

What remains unconfirmed

• Exact timing and capability of Gemini 4 relative to rivals.
• Whether further senior research exits follow the structural change.
• Discovery Loop’s funding size, valuation and first external customers.
• How much of Hassabis’s calendar actually shifts from models to pure AGI strategy and Isomorphic.

The deeper consequence is institutional. Google is no longer asking one Nobel-winning founder to be both product CEO and long-range AGI philosopher. It is asking a 13-year DeepMind builder to ship, and giving the founder permission to chase the problems that animated him in the first place. Dean’s lab adds a third vector that Google still partially owns. Whether that design outruns the talent market and the coding-lag clock is the test the next few model cycles will answer.

Alphabet Now Runs Three Parallel AI Tracks

The org chart that emerged Wednesday is easier to read as three tracks than as a single succession story. Each track has a different clock and a different customer.

• Product and models: Kavukcuoglu ships Gemini, frontier research, the app and developer surface on a consumer and enterprise clock.
• AGI strategy and health: Hassabis advises on models while he drives Isomorphic and long-horizon safety questions on a scientific clock.
• External science loops: Dean, Ghemawat, Vinyals and Quoc Le run Discovery Loop as a public benefit corporation, with Google as founding investor and Cloud partner for the first year.

The three tracks share compute, collaboration paths and brand gravity. They do not share the same decision cadence. That separation is the design choice: product velocity no longer waits on AGI philosophy, and AGI philosophy no longer competes daily with app release trains.

Talent that might have left for rivals can still find a home on one of the three tracks. Whether that home is attractive enough is the open recruitment question the internal executive already flagged.

Isomorphic Labs Draws the Health Mandate

Hassabis keeps the Isomorphic CEO seat and has said he wants to accelerate its push against diseases such as cancer. The staff note framed human health as the first place AI should prove unequivocal value. The Chair and Chief Scientist titles free calendar for that mandate without cutting his advisory rights on DeepMind models.

AlphaFold already supplied the proof of concept that protein work and Nobel-level science can come from the same research culture. Isomorphic is the applied continuation of that culture under a founder who no longer has to clear Gemini product decisions. The mechanism is simple: less executive surface area, more disease-target surface area.

Pichai’s note cast the same move as full attention on shaping AGI’s future. Health applications and AGI safety questions therefore share the founder’s renewed focus. The company still owns the full stack he praised, from TPUs to the Gemini app’s installed base, while he spends more of his week on the problems that first drew him into the work.

Hassabis closed his staff note by saying the best is yet to come. For one day at least, the market priced the opposite. The org chart now matches the work he actually wants to do.