📰 Key Takeaways

Google’s top AI researchers keep leaving, latest as Jonas Adler and Alexander Pritzel join Anthropic, both key to Google’s Gemini model development. This brain drain is becoming a trend that worries Google: just last week, veteran researcher Noam Shazeer with 25+ years announced he’s moving to OpenAI. Shazeer joined Google in 2000, left for three years to start controversial chatbot startup Character.AI, but Google later bought the company for $2.7B partly to bring him back to lead Gemini development. Following that, Google DeepMind Research Director John Jumper also announced he’s leaving for Anthropic. Jumper co-developed AlphaFold with DeepMind CEO Demis Hassabis and won the 2024 Nobel Prize in Chemistry — AlphaFold predicts protein 3D structures from amino acid sequences. Analysts say as OpenAI and Anthropic prepare for IPO, stock incentives will keep attracting top talent, and Google’s talent drain pressure won’t ease anytime soon.

💬 JudyAI Lab Perspective

Google’s top researchers are leaving in waves — Nobel laureate Jumper, Gemini core members Adler and Pritzel moving to Anthropic, legendary engineer Shazeer joining OpenAI — the AI talent landscape is restructuring fast. This isn’t just personnel changes; it’s a structural redistribution of where model capability ceilings sit.

The key detail in this wave: Google even paid $2.7B to acquire Character.AI, the startup Shazeer founded, to keep him — yet it couldn’t stop the broader exodus. When Anthropic and OpenAI are prepping for IPO, stock incentives pull way harder than any salary retention package. From our view, research talent density has always been a leading indicator of model capabilities — Anthropic just snagged both the AlphaFold architect and Gemini core developers in short order; that talent thickness will eventually show up in model performance gaps.

When choosing AI infrastructure, factor in each company’s research team talent movements — that often reveals future capability curves earlier than any press release.

📅 Source Info

🔗 Further Reading