Google’s AI business is booming, so why are DeepMind’s biggest names quitting?

Google’s artificial-intelligence business has rarely looked stronger commercially, yet some of the researchers who helped create that advantage are leaving.

The Gemini app now has 950 million monthly active users, while Google Cloud revenue jumped 82% in the second quarter and backlog reached $514 billion.

Alphabet also raised its 2026 capital-spending forecast to $195 billion-$205 billion.

Against that growth, the exits are striking.

Gemini co-lead Noam Shazeer has joined OpenAI, Nobel laureate John Jumper is moving to Anthropic, and Jeff Dean, Sanjay Ghemawat, Oriol Vinyals and Quoc Le have left to build Discovery Loop.

Demis Hassabis has also stepped away from DeepMind’s daily management, handing operational control to Koray Kavukcuoglu.

DeepMind is losing talent that is difficult to replace

These are not routine departures from a company with thousands of engineers.

Shazeer co-authored the transformer research that underpins modern large language models and later co-led Gemini.

Jumper helped create AlphaFold and shared the 2024 Nobel Prize in Chemistry with Hassabis and David Baker.

Dean spent 27 years at Google and helped build much of the computing infrastructure behind its search and AI systems.

His Discovery Loop co-founders also carry deep research pedigrees across Google Brain and DeepMind.

Wedbush analyst Dan Ives captured the concern after Jumper’s move. “Losing John is a big loss for Google and there is no way to sugarcoat it,” he told Barron’s.

D.A. Davidson analyst Gil Luria separately told MarketWatch that the exits raised concern Google was “losing the war for talent at the frontier of AI.”

The issue is concentration as Frontier AI depends on a relatively small group of people capable of influencing model architecture, training methods and scientific direction.

Talent war offers rewards Google cannot easily copy

The departures are not, by themselves, evidence that Google's technology is failing.

OpenAI, Anthropic and new startups can offer senior scientists autonomy, concentrated computing resources and enormous equity upside.

For researchers who have already built careers inside Google, founding or joining a smaller laboratory can provide a different kind of opportunity.

Jefferies analyst Brent Thill described the situation as an intensifying AI talent war.

He argued that IPO-bound frontier labs are increasingly winning bids for elite researchers, while Google still benefits from deep AI expertise and accelerating Cloud growth.

Discovery Loop illustrates the nuance. The startup is not a clean break with Google: Alphabet is an investor and Google Cloud is a computing partner.

That makes the story less about employees fleeing a sinking ship and more about a labour market in which exceptional researchers can suddenly become founders with access to huge pools of capital.

Google is shifting DeepMind's centre of gravity

Google’s response has been to reorganise rather than retreat.

Hassabis is becoming Alphabet chief scientist and Google DeepMind chair, allowing him to focus heavily on artificial general intelligence, science and Isomorphic Labs.

Kavukcuoglu will run day-to-day operations, including frontier models and Gemini.

The Financial Times reported that operational power is shifting towards Silicon Valley while Google co-founder Sergey Brin has become a stronger strategic voice in the company’s AI effort.

The restructuring suggests Google wants to put clearer operational control around Gemini and its frontier models as competition with OpenAI and Anthropic intensifies.

That may be exactly what Google needs as competition intensifies.

Its commercial numbers show AI adoption and monetisation accelerating, but revenue, chips and distribution do not guarantee leadership at the research frontier.

Google can replace almost any employee. The harder question for investors is what it means when so many people who helped create its AI advantage decide their next breakthrough is better pursued somewhere else.