Google DeepMind's talent problem is no longer a quiet recruiting story. When Noam Shazeer and John Jumper left in the same week, investors treated it as a warning about Google's place in the AI race.
The week of June 17 gave Alphabet the kind of headline you don't want attached to a company trying to prove it still leads artificial intelligence. Noam Shazeer, a co-author of the 2017 Attention Is All You Need paper and a Gemini leader, left Google DeepMind for OpenAI. Two days later, John Jumper, who shared the 2024 Nobel Prize in Chemistry with Demis Hassabis for AlphaFold, said he was leaving for Anthropic.
Alphabet shares fell 5% on June 22, and MarketWatch reported that the company lost $225 billion in market value during the session, its largest one-day market-cap loss on record, according to Dow Jones Market Data. The New York Post put the intraday damage higher, at roughly $270 billion when the stock was down as much as 7%. Use the stricter number and the point is still plain: Wall Street saw two departures as more than ordinary staff turnover.
It should. Shazeer is not just another senior engineer. Google paid about $2.7 billion in 2024 to license Character.AI technology and bring him and other staff back into the company, according to earlier reporting on the deal. Less than two years later, he has gone to OpenAI. If you run a startup, you know what that says about retention. Money can buy a return ticket. It can't always buy commitment.
Jumper's move cuts differently. AlphaFold is one of DeepMind's cleanest claims to real scientific importance, not another chatbot demo or benchmark trophy. Business Insider reported that Jumper is heading to Anthropic after nearly nine years at DeepMind. Bloomberg later reported that two Gemini researchers, Jonas Adler and Alexander Pritzel, were also expected to join Anthropic. Google said talent movement is normal in the industry, and it is. But four names attached to Gemini, AlphaFold and Anthropic in one news cycle is not a normal week.
The pressure inside DeepMind is easy enough to understand. Google needs Gemini to perform in the places customers now notice most, and coding is one of them. Anthropic has turned Claude into a serious developer tool, not just a model family with good safety branding. If Google's own researchers believe the most interesting frontier work is happening at OpenAI or Anthropic, you don't fix that with a memo about culture.
Frankly, the old Google advantage was never only compute. It was the feeling that the best researchers could work on hard problems before the rest of the industry knew how to value them. DeepMind had that aura for years. AlphaFold proved it. The Transformer paper proved it from the broader Google research side. Once the job starts to feel like catching Claude Code on a product timetable, a certain kind of researcher will look for the door.
Google still has one advantage its rivals would love to own. Anthropic itself signed a multibillion-dollar Google Cloud deal in October 2025 for access to up to one million TPUs, with more than a gigawatt of compute expected by 2026, according to the Associated Press. That is an odd but important fact. Anthropic is hiring from Google while also renting part of Google's machine room.
Don't write off that infrastructure. Google's TPUs, data centers, search distribution, YouTube reach and Android footprint are assets OpenAI and Anthropic can't recreate quickly. A researcher can leave Mountain View with knowledge, relationships and taste. They can't take two decades of query history or Google's physical AI infrastructure with them.
But infrastructure is not enough by itself. You still need the people who know which model decisions matter, which benchmark gains are real, and which promising direction is just a costly detour. The uncomfortable question for Google is not whether it can afford to compete. It can. The question is whether the people it most needs still believe DeepMind is the place where the best work will happen.
That is why this story is still current, and why the stock reaction was not simply panic. Investors were not pricing in the loss of two employees. They were pricing in doubt about whether Google's AI research culture can survive the commercial urgency now being placed on it. If Gemini catches up in coding and keeps improving, the current alarm will look overdone. If it doesn't, Shazeer and Jumper will look less like departures and more like early evidence.
Also read: Suno's Spark artist grants come with a permanent silence clause that says more about the lawsuit than the music • AI coding agents are turning code review into the next startup risk • Micron Technology has become the AI infrastructure bet that most investors are sleeping on
