The 2024 Nobel Chemistry laureate left DeepMind.
Nine years in, after building AlphaFold — one of science's greatest achievements — he walked out. Alphabet's stock dropped 6% in five trading days.
But isn't DeepMind the best lab in the world?
Let's talk about what AlphaFold actually is. Predicting the 3D structure of proteins had been one of biology's unsolved problems for over 50 years. It used to take years in the lab to figure out even a single protein structure. AlphaFold solved it with AI.
Today, over 2 million researchers across 190 countries use a database of more than 200 million protein structures it generated. That's why the 2024 Nobel Prize in Chemistry went to both Demis Hassabis and John Jumper.
By that logic, DeepMind should be every AI researcher's dream workplace. But the data tells a different story. DeepMind engineers are 11 times more likely to leave for Anthropic than the other way around, according to one analysis.
Just look at the first half of 2026: Noam Shazeer (Gemini co-lead, co-author of the transformer paper) to OpenAI. John Jumper to Anthropic. Jonas Adler and Alexander Pritzel (key Gemini contributors) with Bloomberg reporting they're also headed to Anthropic. That's six named DeepMind researchers gone or going since February.
So why do they keep leaving?
Three structural forces overlap here.
First, an asymmetry in equity. Google is a public company. What it can offer new researchers is RSUs priced at market rates. Anthropic, on the other hand, is a pre-IPO startup valued at $965 billion. Pre-IPO equity upside structurally can't be matched by public company RSUs.
Second, a gap in research autonomy. DeepMind is a division of Google. The tension between a culture that breeds fundamental science breakthroughs and the priorities of Google's overall product roadmap is structurally unavoidable. Anthropic's 80% two-year retention rate suggests it's not just about money — it signals genuine research autonomy.
Third, the next frontier in science AI is clearly emerging somewhere else. The timing of Jumper's move is interesting. He announced the departure on June 19. Anthropic launched its Claude Science platform on June 30. That's probably not a coincidence.
Why did Anthropic want him specifically?
Anthropic CEO Dario Amodei's October 2024 essay Machines of Loving Grace laid out a vision: compress 50–100 years of biology R&D into 5–10 years. The first half of 2026 saw that vision start executing.
- February: Partnerships with Allen Institute + HHMI (Janelia) announced. Claude integrated directly into live research pipelines
- May: Bristol Myers Squibb deployed Claude across 30,000+ employees
- June 30: Claude Science launched in beta — 60+ scientific capabilities covering genomics, protein structure, drug chemistry
The man who built AlphaFold joining this pipeline signals that Anthropic is treating AI science research not as a feature, but as a core business.
| DeepMind's approach | Anthropic's approach | |
|---|---|---|
| Goal | Fundamental science breakthroughs | Compress R&D timeline by 10× |
| Monetization | Internal use within Google | Direct partnerships with BMS, research institutes |
| Equity | Google RSUs (public company) | Pre-IPO stake ($965B valuation) |
| 2-year retention | Not disclosed | 80% (industry best) |
What you can do right now
If this story isn't just spectator sport for you, here's what to actually do. Claude Science is currently in beta — available on Claude Pro, Max, Team, and Enterprise plans.
- Access Claude Science
Check for the Claude Science tab in Claude.ai. Available immediately if you're on Pro or above. Supports macOS and Linux, and connects to HPC clusters or remote servers via SSH. - Explore the 60+ science skills
Genomics, protein structure, drug chemistry, single-cell RNA sequencing — over 60 preset capabilities. Start with whatever's closest to your research area. - Monitor the AI for Science grant program
Anthropic has been supporting up to 50 projects with $30,000 in credits through the AI for Science program. Follow their official channels for the next round. - Consider team adoption
If your org does life sciences, healthcare, or materials research, it's time to seriously evaluate Claude Enterprise at scale — like BMS did with 30,000+ employees. - Read talent movement as market signal
Where the best researchers go is where the next AI battle is being fought. The DeepMind→Anthropic flow is the market voting that science AI is the next major category.
What we don't know yet
Jumper's title and team at Anthropic haven't been disclosed. The reported Coefficient Bio acquisition is also unconfirmed. Claude Science is in beta with features being added rapidly — check the official page for current capabilities.
Go deeper
Machines of Loving Grace Dario Amodei's original essay on Anthropic's vision for biology AI — the basis for the "10× compression" claim anthropic.com
Claude Science AI Workbench official announcement Full breakdown of 60+ science features, partnerships, and how to use it anthropic.com
Nobel laureate John Jumper is leaving DeepMind for rival Anthropic Original TechCrunch report — the immediate fact record techcrunch.com
AlphaFold Database 200M+ protein structures predicted by AlphaFold — see the research it enables directly ebi.ac.uk
Google's AI Brain Drain Structural analysis of the DeepMind talent exodus. Source of the 11× SignalFire data alphamatch.ai
Anthropic Is Hiring Biologists, Building Wet Labs Detail on the wet lab setup and partnership specifics synbiobeta.com
