Google's real AI problem isn't models — it's a $965B poaching magnet

Demis Hassabis is right: Google has the deepest research bench in AI. He also just watched four of its most decorated people walk out in a single week — including the man who co-wrote the paper that made all of this possible. Both statements are true. Only one of them moved the stock.

Count them. Noam Shazeer, co-author of Attention Is All You Need, to OpenAI (announced 18 June). John Jumper, DeepMind director and 2024 Nobel chemistry laureate, to Anthropic (20 June). Then Bloomberg on 24 June: Jonas Adler and Alexander Pritzel, also to Anthropic. Add Arthur Conmy, who posted that he is joining Anthropic on safety. That is four confirmed exits — three of them straight to Anthropic — inside seven days.

The tempting story is bruised egos and lab drama. The boring, correct story is a cap table. Anthropic raised $65 billion at a $965 billion valuation — the world's most valuable private AI company, ahead of OpenAI — and both labs are circling public listings. A Nobel laureate does not switch teams for a bigger nameplate. He switches for pre-IPO equity you cannot manufacture inside a company whose stock already trades.

There's a friction layer too, and it's petty in the way real workplaces are petty: compute allocated to one of Shazeer's projects was reportedly reassigned to a London team before he left. Stack that on Anthropic's expansion into coding, health and science — exactly where these four are strongest — and the surprise isn't that they left. It's that Google assumed prestige alone would keep them.

Investors didn't wait for a think-piece. Alphabet fell about 5-6% on 22 June, tied to AI-spend fears and — the part that stings — retention risk. That is the market saying the quiet thing out loud: it will forgive a spending binge before it forgives the sense that your best people would rather build the future somewhere else.

Hassabis's rebuttal — biggest bench, wins its share — is not wrong, and it's the only sane thing to say from the podium. But 'we have more of them than anyone' is a strange comfort when the ones you're losing are the ones everyone can name. Depth protects you on average. It does nothing for you on the margin, and the margin is where frontier races are won.

And here's the uncomfortable second-order effect nobody at Google wants to say out loud: talent loss compounds. The first star who leaves makes the second departure easier to justify and the third easier still, because the people who stay start asking why they're the ones holding a public-market stock while their peers cash in on a private rocket ship. Recruiting works the same way in reverse — the names you lose are the names you used to close candidates with. Lose Shazeer and Jumper and your next offer letter is a little less persuasive. None of this shows up in a quarterly metric, which is exactly why it's dangerous: by the time attrition is visible in Gemini's roadmap, the compounding has already happened. Google can afford to lose four people. What it can't afford is for four to become the number everyone remembers.