When the Nobel Prize committee awarded John Jumper the 2024 Nobel Prize in Chemistry for protein structure prediction, they weren't just recognizing AlphaFold — they were validating nearly a decade of Google's AI bet. That bet just lost its star player.
Jumper announced this week that he's leaving Google DeepMind after nearly nine years to join Anthropic. The timing is notable: this is the second high-profile departure from Google's AI division in as many months, and it signals something the market has been pricing in for a while.
It's not about the money
Here's what gets overlooked in the talent migration coverage: these aren't mercenary moves. Jumper could have stayed at Google DeepMind and commanded any salary. The fact that he chose to leave — and leave for a company that's smaller, younger, and has less compute — tells you something about where the momentum is.
The question isn't why they're leaving Google. It's what Google stopped building that made them want to leave.
This isn't a isolated case. The co-occurrence data is stark: Anthropic and Google DeepMind appearing together in articles has spiked 3x in the past month. That's not coincidence — that's narrative forming in real-time.
What Google gets wrong
Google still has the most researchers, the most compute, and the most data. What it doesn't have is a clear story about what it's building next. DeepMind was AlphaFold and Go and protein folding — domains where being first mattered. But the current race isn't about domains. It's about general capability, and that's not a problem you solve with a research lab that's geographically and culturally separate from the product team.
Anthropic doesn't have that problem. It's small enough that everyone is aligned on the same mission. That's worth more than $3.2 billion in TPU credits.
The talent flow is a leading indicator. When the best researchers vote with their feet, they're not just changing employers — they're changing which narrative they believe. Right now, the bet is on the smaller company that moves faster.