Our Take
Google built the talent pipeline that now funds its competitors; public markets will make the hemorrhage worse before it stops.
Why it matters
When a research lab loses four senior scientists in three weeks, it is losing institutional knowledge, model development velocity, and credibility with the next generation of PhDs. The timing matters: Anthropic and OpenAI going public will make equity packages irresistible to researchers who spent years building Gemini under Google's internal constraints.
Do this week
AI teams at Google: document your Gemini architecture and training decisions in writing before more architects follow Adler and Pritzel out the door; you will need it to onboard replacements or defend your roadmap to leadership.
Four departures in three weeks
Jonas Adler and Alexander Pritzel are joining Anthropic from Google, where they were key contributors to Gemini development. Their departures follow three others announced in the past two weeks: Noam Shazeer to OpenAI (after 26 years at Google, interrupted only by his Character.AI startup), and John Jumper to Anthropic (director of Google DeepMind and co-winner of the 2024 Nobel Prize in Chemistry for AlphaFold work).
The timing is not accidental. Shazeer announced his move last week; Jumper followed days later. Anthropic and OpenAI are both preparing public offerings, which means equity packages tied to valuations that will dwarf anything Google can offer without triggering internal equity adjustments and board fights.
Talent concentration becomes a liability
Google paid $2.7 billion to acquire Character.AI (effectively an acqui-hire of Shazeer) to retain him for Gemini work. That investment is now walking out the door to a competitor. Jumper's departure is worse: he is not just a researcher but a lab director with hiring authority and roadmap influence at DeepMind. When directors leave, so do their reports.
The structural problem for Google is asymmetric. Anthropic and OpenAI are moving to public status, which lets them offer meaningful equity upside tied to company value. Google's equity is already mature; researchers see more absolute upside in an earlier-stage public company that might 10x than in Alphabet stock that might appreciate 15% annually. Add in Google's internal politics (Gemini faced well-documented pressure to manage model behavior in ways researchers may not have chosen freely) and the pull becomes irresistible.
For Google's AI product roadmap, this is a reliability problem. Gemini competes on capability and speed. Both depend on continuity in the teams that understand the model's training data, architecture choices, and failure modes. When you lose Adler and Pritzel, the people who built those systems, you lose months of catch-up time for whoever replaces them.
What this means for the field
If this trend accelerates, Google's ability to field competitive models may lag behind OpenAI and Anthropic by 2027. The company will likely respond with aggressive counter-offers and internal promotions, but neither fixes the core problem: Google's organizational structure and public-company constraints make it a less appealing place to do frontier research than a well-funded but younger rival.
For practitioners, the lesson is simple: the lab with the best researchers ships better models. Track these departures. They signal where the next capability jumps are likely to happen.