Google Lost Four Senior AI Researchers in a Week. Two of Them Built the Things Google Runs On.

Noam Shazeer told the world he was leaving Google for OpenAI on June 18. The next day, John Jumper said he was leaving Google DeepMind for Anthropic. Bloomberg then reported that two more Gemini researchers, Jonas Adler and Alexander Pritzel, were also on their way to Anthropic. That is four senior people walking out the door to direct competitors inside a single week.

The market noticed. Alphabet closed down about 5% on Monday, June 22, its worst day in over a year. The stock fell as much as 7.2% intraday, the steepest move since February, and roughly $225 billion in market value evaporated. A Sunday interview in which Microsoft CEO Satya Nadella warned about a handful of "AI Giants" controlling the market did not help the mood, but the trigger was the departures.

Who actually left

Two of these names are not interchangeable headcount.

Noam Shazeer is a co-author of "Attention Is All You Need," the 2017 paper that introduced the Transformer. Almost every large language model shipping today, including Google's own Gemini and the AI Overviews sitting on top of search, runs on that architecture. He helped build LaMDA, Google's early chatbot, in 2021, then left in frustration and co-founded Character.AI. Google paid to undo that: it brought him back in 2024 through a Character.AI licensing deal reported at $2.7 billion and made him a co-lead on Gemini. Less than two years later he is gone again, this time to OpenAI. Sam Altman welcomed the hire publicly.

John Jumper led AlphaFold at DeepMind. AlphaFold predicts the three dimensional shape of a protein from its amino acid sequence, a problem biologists had chipped at for fifty years, and it has since produced structure predictions for more than 200 million proteins. Jumper shared the 2024 Nobel Prize in Chemistry with DeepMind CEO Demis Hassabis for that work. He is leaving after nearly nine years, taking time off before he starts at Anthropic, which has said it wants to do more in biology.

The number behind the stock move

A 5% drop is the market's reaction, not the story. The story is cadence.

Google announced at I/O in May that Gemini 3.5 Pro was ready for a wide release, targeting general availability in June. It is now expected in July. By the time it ships it will land roughly four months after Google's last frontier-tier release, Gemini 3.1 Pro in February. For comparison, in that same window Anthropic shipped two Claude Opus updates and an entirely new model line, and OpenAI kept moving too. Google's current top models, Gemini 3.5 Flash and 3.1 Pro, frequently sit outside the top five on public benchmark leaderboards, behind models from Anthropic, OpenAI, and Chinese labs like Zhipu AI and MiniMax.

So when investors saw two of the people most associated with Google's technical lead leave in 48 hours, and saw the next model slip, they connected the two. Reasonable. Bloomberg also reported that DeepMind staff have flagged a more specific gap: Google still does not have a strong product for businesses building AI coding tools, which is exactly the ground Anthropic and OpenAI have taken. Sundar Pichai said the quiet part in May, admitting Google was "a bit behind" on agentic coding.

What this does and does not mean

It does not change how Gemini behaves for anyone using it today. A model that scores fourth on a leaderboard is still a very good model, and Google's distribution remains the envy of every lab on that list. Gemini ships to people who never downloaded an app, through Search, Android, and Workspace. That advantage does not evaporate because two researchers changed badges.

What it does change is the read on retention, and that is the part Google should worry about. The company paid a reported $2.7 billion to get Shazeer back, installed him at the top of its flagship model effort, and could not keep him. It has been handing out a special class of fast-vesting stock to stop poaching by Meta and others. The people taking these exits are already rich, which makes money a weak explanation. Fortune's Jeremy Kahn, who has interviewed Jumper repeatedly, argued the simplest read is that Jumper thought the scientific opportunity at Anthropic was genuinely better. If that is true, it is worse news for Google than any pay package, because you cannot fix it with cash.

Shazeer's old critique of Google, that it had grown too bureaucratic and risk averse to move fast, is the thing current and former employees keep repeating. Google can plausibly survive on the "play not to lose" strategy that a company with billions of users and real profits to defend can afford. That strategy just does not tend to keep the people who want to win.

He is also not the only one out. David Silver, one of DeepMind's earliest hires and a foundational figure in reinforcement learning, recently left to start his own company, Ineffable Intelligence. DeepMind used to brag that nobody left. That is no longer the line.