In one week, two of Google’s most important AI researchers left for rivals. Noam Shazeer, co-lead of the Gemini models and co-author of the 2017 “Attention Is All You Need” paper that created the Transformer architecture behind every modern large language model, left for OpenAI on June 18. One day later, John Jumper, who won the 2024 Nobel Prize in Chemistry for AlphaFold, left Google DeepMind for Anthropic. Alphabet’s stock fell as much as 7% on June 22, its worst day in a year, erasing roughly $269 billion in market value.
The detail almost nobody is leading with: Alphabet reportedly owns about 14% of Anthropic, so Google may literally profit from losing its Nobel laureate to the rival. Google paid $2.7 billion in 2024 to bring Shazeer back, and couldn’t keep him two years later. DeepMind once bragged that nobody ever left. The exits land while Gemini sits outside the top five on most benchmark leaderboards, and they crystallize a read the market is now pricing in: the frontier race looks like Anthropic versus OpenAI, and even Google’s own stars are voting with their feet.
Best for anyone tracking the AI race, Alphabet investors, and people who like a story with a genuinely strange twist. Not ideal for readers who want a model review, because this one is about the people, not the product.
Two years ago, Google paid 2.7 billion dollars to bring one man back.
He was Noam Shazeer, and he had helped invent the thing the entire AI industry is built on. Last week he left again.
Two days later, a Nobel Prize winner walked out the same door. And the strangest part of the whole story is that Google might come out ahead on one of them.
Who Just Left, and Why It Stings
Noam Shazeer announced on June 18 that he was leaving Google for OpenAI. If you only know one name in AI research, this might as well be it. Shazeer co-authored the 2017 paper “Attention Is All You Need,” the work that introduced the Transformer architecture. Every major large language model today runs on the idea that paper put into the world. GPT, Claude, Gemini, all of them. His Transformer work is the literal foundation under Google’s own AI Overviews and AI Mode.
Then it happened again. One day later, John Jumper announced he was leaving Google DeepMind for Anthropic. Jumper isn’t just a senior researcher. He won the 2024 Nobel Prize in Chemistry for AlphaFold, the system that cracked protein structure prediction and became DeepMind’s most celebrated scientific achievement. He spent nearly nine years there. Now he’s taking the most credible AI-for-science résumé on earth to a competitor.
Two legends. One week. Both to the two labs Google is supposedly racing. As Fortune put it, there was a time DeepMind bragged that nobody ever left. That is very much not the case anymore.
The Market Did the Math Immediately
Investors did not treat this as inside baseball. Alphabet shares fell as much as 7% on June 22, the company’s worst trading day in roughly a year, wiping out around $269 billion in market value in a single session. The stock had held up fine in the days right after Shazeer’s news. Jumper’s exit was the one that broke it.
The reason the market reacted so hard is that these two weren’t working on side projects. Shazeer co-led Gemini, Google’s flagship model line. Jumper anchored the scientific reputation that made DeepMind DeepMind. When the people leading your most advanced work leave for the competition in the same week, investors stop reading it as turnover and start reading it as a signal about who’s winning. D.A. Davidson’s analyst put it bluntly: the race at the frontier right now appears to be between Anthropic and OpenAI. Google wasn’t on that list.
It also lands against an expensive backdrop. Alphabet has committed to $180 billion to $190 billion in AI spending for 2026. It raised $141 billion in debt and equity since October to fund the buildout. Spending that much while your top researchers walk out is the exact combination that makes investors nervous about whether the money is buying results.
The Twist Nobody Is Leading With
Here’s where the story stops being a normal talent-exodus piece and becomes genuinely strange.
Alphabet reportedly owns about 14% of Anthropic. So when John Jumper, a sitting Nobel laureate, leaves Google DeepMind to join Anthropic, Google doesn’t just lose a researcher. It loses him to a company it partly owns. If Anthropic’s value climbs because it just landed the most credible AI-for-science hire in the industry, a chunk of that upside flows back to the very company that lost him. Google may literally profit from its own brain drain.
That’s the kind of contradiction the AI race keeps producing. The frontier labs are so tangled together through investment that “rival” stops meaning what it used to. Google funds Anthropic. Anthropic poaches Google’s Nobel winner. Google’s equity stake goes up. Everyone involved can frame it as a win, which is its own kind of tell about how strange this market has become. We’ve written before about how the lines between these labs blur once the money gets involved, and this is that dynamic at its most absurd.
It doesn’t soften the blow to Google’s reputation, though. Owning a piece of the lab that’s eating your lunch is not the same as keeping your people. The equity goes up. The talent still left.
Why They Actually Left
Money is the easy explanation, and it’s worth taking seriously before dismissing it. Both men stand to make generational wealth if Anthropic and OpenAI go public in the coming months, which both are expected to do. There’s a real difference between being a rich researcher with stock options and being early equity in a lab about to IPO at a near-trillion-dollar valuation.
But money is probably not the main driver, and that’s the part that should worry Google most. Both researchers were already extremely wealthy. Shazeer reportedly made hundreds of millions from the Character.AI deal that brought him back. Fortune’s reporter, who has interviewed Jumper repeatedly, said plainly that Jumper doesn’t read as someone primarily motivated by money, and wouldn’t leave unless he believed the scientific opportunity at Anthropic was actually better. If that’s true, it’s far worse news for Google than a bidding war would be. A competitor outbidding you is a problem you can solve with cash. A competitor offering a better place to do the work is not.
There’s history here too. Before Shazeer left Google the first time, he’s thought to have written an anonymous internal memo, later leaked, criticizing the company for being too bureaucratic, slow, and risk-averse to win in AI. ChatGPT’s launch shortly after seemed to prove him right. He came back in 2024. Then he left again. When the same person diagnoses the same disease twice and leaves twice, it stops looking like a personality clash and starts looking like a recurring condition.
What This Means for the AI Race
Step back and the picture sharpens. Google’s own models, Gemini 3.5 Flash and Gemini 3.1 Pro, frequently rank outside the top five on AI benchmark leaderboards, trailing not just Anthropic and OpenAI but Chinese labs like Zhipu AI and MiniMax. The talent leaving and the benchmarks slipping are the same story told two ways: the lab that invented the Transformer is no longer the obvious leader in using it.
None of this changes how Gemini or AI Overviews behave for you today. Google is still enormously profitable, still reported 82% earnings growth last quarter, and still has most of Wall Street rating it a buy. The products work. The search dominance is real. This is a story about trajectory, not a collapse.
But trajectory is what the frontier race runs on. Anthropic has an AI-for-science event scheduled for June 30, where Jumper’s hire will almost certainly take center stage. OpenAI has filed for its IPO. Both are hiring the people who built Google’s lead, partly with money Google helped provide. The open question Google has to answer isn’t whether it can build great AI. It already did. It’s whether it can hold onto the people who know how, and right now the answer is leaving for the competition two at a time.
Google paid 2.7 billion dollars to bring the inventor of the Transformer home. He left anyway. That number isn’t the cost of one hire. It’s the price of finding out that the most important thing in AI isn’t the model or the money. It’s the people, and they can always walk.
