Google's AI Chief Scientist Jeff Dean Exits After 27 Years as DeepMind's Hassabis Steps Back

Google spent the last two weeks proving it can print money from AI and losing some of the people who made that possible.

On the earnings call last month, Alphabet CEO Sundar Pichai told investors that 90% of Fortune 100 companies now use Gemini Enterprise, according to CNBC. Google Cloud posted 82% revenue growth, a number that dwarfs the pace at Amazon and Microsoft. The stock is up 16% this year, on top of a 65% run in 2025 that beat every other megacap.

Then on Wednesday, Google announced chief scientist Jeff Dean is departing after 27 years at the company, and Demis Hassabis is stepping down as CEO of Google DeepMind to become chairman of the unit, CNBC reported. Alphabet shares had already fallen after the earnings report on concerns about capital spending, and dipped again after the leadership news.

Dean is not leaving alone. CNBC reported he's departing alongside other Google researchers, including Sanjay Ghemawat, longtime engineering partners who helped build core infrastructure behind Google's search and AI systems for decades.

The Split Google Faces

The tension is structural. Building frontier AI models means enormous upfront spending on compute and research with no guaranteed payoff. Selling cloud services and enterprise AI tools is proving to be a far more efficient, faster-growing business right now.

Google holds the patent lineage that matters here. The 2017 transformer paper, authored by Google researchers, is the technical foundation underneath the entire generative AI boom, including the large language models built by OpenAI and Anthropic. Google invented the blueprint. Other companies have been faster at turning it into products people want to buy.

Tomasz Tunguz, founder of Theory Ventures, told CNBC that companies may not need frontier-level models to satisfy most business demand anymore. "I think we are at that place with AI, particularly for a lot of white-collar work, where many of the models that are reasonable are good enough," Tunguz said. He added that the next generation of cutting-edge models will likely matter most in specialized, compute-heavy domains, not everyday office work.

Google's cloud business doesn't need to win the model-quality arms race to keep growing. For researchers who joined Google to chase frontier breakthroughs, the pitch to fine-tune sales strategies is a tougher sell.

Why Researchers Are Leaving

According to people familiar with the matter who spoke to CNBC on condition of anonymity, some Google researchers have grown frustrated watching the company sell its homegrown TPU chips, tensor processing units that compete with Nvidia's GPUs, to outside customers including Anthropic, while feeling starved of the computing capacity they need for their own ambitious projects.

Layers of internal approval required to move research into shipped products is a recurring complaint inside Google, CNBC reported. That bureaucracy is exactly what makes smaller, faster-moving shops like OpenAI and Anthropic attractive to researchers who'd rather build than navigate a corporate chain of command.

This isn't a new problem for big tech. Any company with Google's market cap, roughly $4 trillion, has to answer to shareholders who care about quarterly cloud revenue more than they care about a research team's ten-year moonshot. Google isn't doing anything wrong by running cloud like a business. But the company is competing against startups that don't have to justify every compute allocation to a board.

What's Unresolved

No successor to Dean as chief scientist has been named in the reporting available. It's also unclear whether Hassabis moving to chairman signals a broader restructuring of DeepMind's leadership or a one-off transition tied to his specific role.

CNBC's framing treats this largely as a talent and strategy story inside Google, which is fair as far as it goes. What's less clear from that coverage is how Google's TPU sales to competitors like Anthropic square with keeping its own research teams adequately resourced. If Google is renting out its best chips to the same labs poaching its scientists, that's a decision investors and employees alike deserve a clearer explanation for.

The next earnings call will be the place to watch. If capital expenditure concerns that already dented the stock resurface alongside more departures, Wall Street's patience with Google's dual-track strategy, dominant cloud business, and uncertain research retention may not last forever.