One Week, Four Departures

Between June 18 and June 24, 2026, Google’s AI organization lost four of its most prominent researchers to direct rivals. Noam Shazeer — a co-inventor of the Transformer architecture and co-lead of Gemini — announced he was joining OpenAI on June 18. The next day, John Jumper, who shared the 2024 Nobel Prize in Chemistry for AlphaFold, said he was leaving DeepMind for Anthropic after nearly nine years. By June 24, Bloomberg reported that Jonas Adler and Alexander Pritzel, two senior researchers who contributed to both Gemini and AlphaFold, were also heading to Anthropic.

Four researchers. Six days. Google’s stock fell roughly 6% over three trading sessions following the first two announcements — about $22 per share erased. Alphabet’s worst single-week AI talent loss on record became immediate news across the industry.

Who Left and Where They Went

Noam Shazeer → OpenAI

Shazeer co-authored “Attention Is All You Need,” the landmark 2017 paper that introduced the Transformer architecture underlying virtually every modern language model, including GPT, Gemini, and Claude itself. After leaving Google in 2021 to co-found Character.AI, he was pulled back in 2024 when Google paid roughly $2.7 billion to license Character’s technology — an acqui-hire intended to anchor Gemini development. Less than two years later, he has left again. At OpenAI, Shazeer will serve as Lead for Architecture Research, working on next-generation model design. The irony of Google paying billions to reacquire someone who then joined the chief rival is not subtle.

John Jumper → Anthropic

Jumper spent nine years at DeepMind, most of them leading the AlphaFold program. That effort earned him a share of the 2024 Nobel Prize in Chemistry alongside Demis Hassabis and David Baker — the first Nobel ever awarded largely for work done by AI. AlphaFold has now predicted the structure of more than 200 million proteins, accelerating drug discovery pipelines at pharmaceutical companies, universities, and research hospitals worldwide. Jumper announced his move to Anthropic on X on June 19. His specific role hasn’t been disclosed, but for a company that frames itself as a safety-focused research lab rather than a product company, his arrival brings scientific legitimacy that no benchmark score can provide.

Jonas Adler and Alexander Pritzel → Anthropic

Adler worked on Google’s AI coding efforts; Pritzel focused on pretraining — the foundational stage where large models learn from enormous datasets. Both also contributed to AlphaFold alongside Jumper. Bloomberg’s June 24 report said both are expected to join Anthropic, adding a third and fourth departure to what was already a damaging week. Their moves suggest Jumper’s exit may have been a catalyst rather than an anomaly — a departure that made it easier for nearby colleagues to act on intentions they had been weighing for some time.

What Google Is Actually Losing

These are not peripheral contributors. Shazeer’s 2017 paper didn’t just influence deep learning — it defined the decade. His subsequent work on Mixture of Experts architectures shaped how frontier models are structured today. Losing him to OpenAI gives rivals direct access to the architectural thinking that drove Gemini and potentially years of institutional knowledge about what works and what doesn’t at scale.

Jumper, Adler, and Pritzel represent the AlphaFold core. The concern isn’t that their work will now benefit Anthropic’s drug discovery efforts — Anthropic isn’t a life sciences company. The concern is subtler: Google is losing people who know how to do research at the highest level. The discipline required to build AlphaFold — rigorous, multi-year scientific work with verifiable real-world results — is precisely the intellectual culture that frontier AI labs need more of, not less.

The collective damage to the Gemini program is also concrete. Shazeer co-led it. Pritzel contributed to pretraining. Architecture understanding and training intuition walk out the door with them. Training runs are expensive; rebuilding institutional knowledge is slower and costlier still.

Why Rivals Are Winning the Talent War

The financial pull is real. Anthropic’s $65 billion fundraising round closed in early 2026, and the company has reached an annualized revenue run rate approaching $4.7 billion — growth of more than 80x year-over-year. That kind of trajectory attracts top talent both on compensation and on conviction that the work will matter. OpenAI, with its $122 billion valuation and IPO filing, offers similar signals of momentum. When a researcher believes their work has a better chance of reaching the world through Company B than Company A, the compensation premium required to stay at A gets significantly larger.

Organizational velocity also matters. Google runs AI development inside a company generating hundreds of billions in search advertising revenue annually. The organizational caution required to protect that business creates friction that smaller, purpose-built labs don’t face. Multiple departing researchers noted — on background, according to Fortune and Axios reporting — that the ability to move faster, take more research risk, and operate without the constraints of product commercialization timelines was a real factor in their decisions.

For Jumper specifically, there may also be a mission alignment at work. Anthropic was founded explicitly around AI safety research, with a long-term focus on understanding how advanced AI systems behave and how to make them safer. For a scientist who spent a decade doing careful, rigorous work in protein structure prediction, a lab that directs that rigor toward AI’s most fundamental questions may be more compelling than one running on a product clock.

What Comes Next for Google’s AI Program

Google isn’t finished. Demis Hassabis, who also holds a Nobel Prize, remains and leads DeepMind’s research agenda. Gemini 3.5 Flash, released in early June, is technically competitive at the frontier. Google still has data advantages, compute infrastructure, and research teams that no startup can replicate in the near term.

But the talent narrative now has its own momentum. If another high-profile departure follows before year-end — and there is no particular reason to think the wave has stopped — it stops being a bad week and starts becoming a structural story. Alphabet’s board needs to answer not just to shareholders but to the researchers who remain: why is DeepMind the right place to do the most important AI work of the next decade? That answer was easier to give when the top of the org was stable and the models were clearly leading on benchmarks.

The week of June 18 was a real blow — not fatal, but not ordinary either. In the AI talent market, perception compounds. The best signal that a lab is the place to be is that the best people keep choosing it. Right now, four of Google’s best chose somewhere else.

Further Reading

• Nobel laureate John Jumper is leaving DeepMind for rival Anthropic — TechCrunch’s original report on Jumper’s departure and his nine years at DeepMind.
• As top talent leaves Google DeepMind, some question if the lab can remain at the forefront — Fortune’s deeper analysis on what the departures mean for Google’s long-term AI competitiveness.
• Google Poised to Lose Two More High-Profile AI Staffers to Anthropic — Bloomberg’s June 24 reporting on Adler and Pritzel’s expected departures.