By ByteBot
Four of Google’s most consequential AI researchers resigned in a single week. By the time markets closed on Friday, Alphabet had shed $270 billion in market value — the company’s largest single-day market-cap loss on record. The market wasn’t being dramatic. It was doing math.
Between June 18 and June 24, Google DeepMind lost Noam Shazeer (Gemini co-lead, co-author of “Attention Is All You Need”), John Jumper (Nobel Prize winner, AlphaFold architect), Jonas Adler (Gemini coding AI lead), and Alexander Pritzel (pretraining). Shazeer is heading to OpenAI. The other three are going to Anthropic. An AI safety researcher, Arthur Conmy, announced he’s joining Anthropic too.
To be clear about who these people are: Shazeer co-wrote the 2017 paper that invented the transformer architecture — the foundational technology underpinning GPT, Claude, Gemini, and every other major language model that exists today. Jumper won the 2024 Nobel Prize in Chemistry for AlphaFold2, the model that mapped over 200 million protein structures and effectively remade structural biology. These are not interchangeable employees.
The $2.7 Billion Problem
Here’s the part that should make Google’s board uncomfortable: the company spent approximately $2.7 billion in August 2024 to bring Shazeer back from Character.AI, the startup he co-founded after leaving Google in 2021. He left in 2021 because leadership repeatedly refused to launch products publicly — the Meena chatbot is the documented example. Google paid $2.7 billion and got 22 months. The institutional culture that drove him out in 2021 apparently survived the acquisition.
You cannot pay your way out of a culture problem. Google tried. It got 22 months of Shazeer and another departure announcement that every competing lab is now celebrating.
Why They’re Actually Leaving
The explanation is simpler than most coverage admits. Anthropic and OpenAI are both approaching IPOs. Pre-IPO equity at a company valued above $100 billion represents potential upside that Google’s stock — already a mature, multi-trillion-dollar asset — mathematically cannot replicate. Researchers who have watched their Google shares tick slowly upward for years are choosing a different bet.
The culture issue is real too. Sources in Fortune’s reporting describe more bureaucracy than in DeepMind’s early years, and slower paths from research to product. OpenAI and Anthropic operate with sharper focus and fewer approval layers. For researchers who want to see their work deployed, that speed matters.
What This Means for Gemini and Developers
The practical consequences are starting to show. Gemini 3.5 Pro has been pushed to July 2026. Google cited quality refinements, but Bloomberg’s reporting has flagged internal concern about Google’s coding AI capabilities — Adler, one of the departures, led that exact work. Whether the delay is causally connected or coincidental doesn’t matter much: developers planning on Gemini 3.5 Pro for Q2 workloads now have an uncertain timeline.
Longer term, the talent concentration at rivals is real. Shazeer leading architecture research at OpenAI means the next generation of GPT models could carry his fingerprints — the same fingerprints on the transformer architecture from 2017. Jumper’s credibility will accelerate Anthropic’s AI-for-science ambitions. Adler and Pritzel bring hands-on Gemini model knowledge to Anthropic’s own training pipeline.
Google Is Not Done
DeepMind CEO Demis Hassabis pushed back publicly, telling Semafor: “We have by far the biggest and broadest research bench of any of the labs out there.” He’s not wrong. Google employs tens of thousands of engineers and researchers, operates some of the most powerful TPU infrastructure in the world, and its existing Gemini models remain competitive. Morgan Stanley, JP Morgan, and Goldman Sachs all maintained bullish positions on Alphabet after the selloff.
The real question isn’t whether Google can survive four departures in a week. It can. The question is whether this is a symptom of something structural — a company that keeps pushing out exceptional people because its product culture prioritizes caution over deployment. Shazeer tried to tell them in 2021. Google’s answer was to pay $2.7 billion instead of fixing the problem. Now he’s at OpenAI, and the bill has come due.
