The departures came in quick succession last week: four of Google’s most senior AI researchers handed in notice within days of one another, according to people familiar with the matter. By the time the week ended, Alphabet Inc. had lost roughly $186 billion in market value, and investors inside and outside the company were asking the same question: is the exodus a run of bad luck or a structural problem?
The identities of the four researchers have not been fully disclosed, but people familiar with the matter describe a mix of DeepMind scientists and Gemini contributors, several of them targets of long courtships by rival labs. The exits follow a wave earlier in the year that already thinned Google’s front ranks: Noam Shazeer, a pivotal architect of the Gemini models, left for OpenAI in June, and two days later John Jumper, the Nobel Prize-winning DeepMind researcher behind AlphaFold, announced he was joining Anthropic. Two more DeepMind researchers, Jonas Adler and Alexander Pritzel, followed Jumper to Anthropic within days.
The August wave has the same shape. The destinations vary — OpenAI, Anthropic, and at least one new startup founded by the departing researcher, according to people familiar with the matter — but the pattern is consistent. Google is losing the people who build its most important products to the companies that compete with them, and it is losing them to equity packages and research freedom that its compensation system struggles to match.
The market reaction is a measure of what investors think those people are worth. Alphabet’s shares fell sharply over the week, wiping out the $186 billion in value, as analysts revised their views on the company’s ability to defend its position in the AI race. Google remains one of the largest and most profitable companies in the world, with a search franchise that prints cash and a cloud business growing at scale, but its AI moat is increasingly identified with specific individuals.
That identification has a history. DeepMind was created in 2010 and acquired by Google in 2014 in a deal whose price — reported at more than $500 million — was largely a payment for a handful of researchers. The merger of DeepMind and Google Brain in 2023 was partly an attempt to reduce internal friction that was driving people out. The departure of Shazeer, who had left once before to found Character.AI and returned in 2024, showed that even a seven-figure retention package does not guarantee loyalty.
The economics of the talent war have changed. OpenAI and Anthropic, both private companies with valuations in the hundreds of billions or higher, can offer researchers equity that compounds with each model release, alongside promises of scientific autonomy that a bureaucracy-heavy Google finds harder to match. Startups offer founding roles. The asymmetry is not in salary, which Google can match, but in upside and control.
Google’s response has been limited. DeepMind chief Demis Hassabis publicly acknowledged Jumper’s departure, a rare public statement about an individual researcher, but people familiar with the matter say he has not yet articulated a structural response — no new equity program for top scientists, no reorganization of how senior researchers are compensated or empowered. The company’s model pipeline has not paused, but the concern among investors is about what the pipeline looks like in three to five years, when the researchers leaving today become the architects of competing systems.
The stakes are visible in the numbers. Alphabet is spending tens of billions of dollars a year on AI infrastructure, and the returns depend on the quality of the models it ships. A single top researcher can be the difference between a model generation that leads the field and one that follows it. When the market prices that dependence, the loss of four researchers in a week moves the stock the way a missed earnings number would.
There is a counterargument, and Google’s defenders make it. The company still employs thousands of AI researchers, has more compute than any lab outside the hyperscalers, and owns distribution for AI products that rivals cannot match. Shazeer’s and Jumper’s departures have not stopped Gemini updates or DeepMind’s research output. Talent flows both ways — Google has hired from OpenAI and Anthropic too — and the most recent departures may have more to do with individual circumstances than with a systemic failure.
The market is not fully buying that case. A company whose stock drops $186 billion on talent news is being valued, at the margin, on its ability to retain a small number of irreplaceable people. That is a fragile basis for a multi-trillion-dollar market capitalization, and it explains why investors are watching Google’s next moves closely: retention packages, organizational changes, or the quiet promotion of new leaders inside DeepMind.
For the AI industry, the week is a preview of the competition’s next phase. The frontier labs have reached rough parity in model quality, and the scarce input is no longer compute alone — it is the people who know how to use it. The winners of the next two years will be the companies that can attract and keep researchers while their competitors try to take them away. Google, which has lost more top researchers in the past three months than any other lab, is the test case.
The company will not comment on individual departures, and the four who left last week have not spoken publicly about their plans. The market has already voted. Whether Google can replace what it lost — with money, with structure, or with the argument that its platform is the best place to do AI research — will be measured in the stock price and in the quality of the models it ships next year.
