The DeepMind Drain: Nobel Laureate Jumper and Gemini Researchers Defect to Rivals

A week that cost Alphabet a small country's GDP

The most consequential AI story of the past few days was not a model launch or a chip. It was a series of resignation posts. Over roughly one week in late June 2026, Google DeepMind — the research lab that arguably invented the modern era of deep learning — watched several of its most decorated researchers walk out the door toward direct competitors. The market reaction was brutal: according to Fortune, Alphabet shares tumbled more than 5% on the Monday following the news, and Crypto Briefing reports the departures wiped roughly $270 billion off Alphabet's market capitalization.

That is an extraordinary price tag for a handful of personnel moves. It tells you that investors no longer see frontier AI as a question of compute and data alone. They see it as a question of who, specifically, is in the building.

Who left, and where they went

The headline departure is John Jumper, who announced on Friday, June 20 that he is leaving after nearly nine years to join Anthropic — though he said he will take time to recharge first. Jumper is not an ordinary researcher: he shared the 2024 Nobel Prize in Chemistry with DeepMind CEO Demis Hassabis for AlphaFold, the system that predicted the structures of over 200 million proteins. In his posted statement, Jumper credited Hassabis for "taking a real chance" on him and called DeepMind "a special place."

He was not alone. According to Fortune, Noam Shazeer — a Gemini co-lead and a co-author of the foundational "Attention Is All You Need" Transformer paper — announced his own departure days earlier, on June 18–19. In his case, the destination was OpenAI, not Anthropic. Shazeer has a history here: he previously left Google over frustration with slow commercialization, co-founded Character.ai, and returned in 2024.

Crypto Briefing adds two more names headed to Anthropic: Jonas Adler, who worked on Google's AI coding efforts, and Alexander Pritzel, who worked on pretraining. Notably, both reportedly contributed to AlphaFold alongside Jumper. Fortune also references David Silver, a pioneering reinforcement-learning researcher and early DeepMind employee, who in recent months left to found a startup. The clustering matters: this is not one star chasing a paycheck, but a coordinated-looking drain of talent across protein science, coding, and core pretraining.

Why this matters more than a normal poaching cycle

Talent moves between AI labs constantly, and Hassabis said as much. Crypto Briefing quotes him stating that "movement between leading laboratories was expected in the current market," and that Google retains "the largest and broadest research team in the industry." That framing is fair and worth holding onto — losing four or five people, however senior, does not empty out a lab of thousands.

But the reason markets flinched is captured in Fortune's reporting on DeepMind's competitive standing. Per that piece, DeepMind's top models — it cites Gemini 3.5 Flash and Gemini 3.1 Pro — were ranking outside the top five on AI benchmarks, trailing Anthropic, OpenAI, and Chinese labs. It also notes a slower release cadence, with a Gemini 3.5 Pro arriving roughly four months after the prior model, while Anthropic shipped multiple Claude updates plus a new model line in the same window. When a lab is already perceived as a half-step behind, losing its most visible scientists reads less like normal churn and more like a vote of no confidence from the inside.

The cultural subtext

The most damaging detail in the coverage is not a number — it's a description. Fortune reports that current and former employees characterized DeepMind as "bureaucratic, sometimes bordering on sclerotic, and highly risk-averse," with one analyst noting the lab is "burdened by its size." That echoes Shazeer's original reason for leaving years ago: that Google moved too slowly to commercialize AI.

This is the real story beneath the stock move. Money is not the obvious lever here — Google can outbid almost anyone. What smaller, faster rivals appear to be selling is velocity and ownership: the ability to ship, to see your research reach users, and to do it without wading through layers of process. For researchers who measure their careers in breakthroughs rather than bonuses, that pitch can outweigh a larger compensation package.

The Anthropic and OpenAI angle

The flip side of DeepMind's loss is a remarkable consolidation of talent at its rivals. Crypto Briefing reports that Anthropic recently raised $65 billion at a $965 billion valuation, describing it as the world's most valuable private AI company — figures worth treating with some caution, as they come from a single secondary source rather than an Anthropic filing, and private valuations are notoriously soft. If accurate, they explain how Anthropic can credibly recruit a Nobel laureate: it is no longer the scrappy underdog but a capital-rich magnet expanding into coding, healthcare, and scientific applications — precisely the domains where Jumper, Adler, and Pritzel built their reputations.

OpenAI, for its part, reclaiming Shazeer is symbolically potent given his authorship of the Transformer paper that underpins essentially every modern large language model. The talent market is becoming a zero-sum scoreboard, and right now Google is on the wrong side of it.

Hype versus reality

A few cautions are in order. First, several of these moves are reported as planned or announced, not necessarily completed — Jumper himself said he is taking time off before starting. Second, the precise financial figures vary by outlet and should be read as approximate: a "more than 5%" drop and a "$270 billion" wipeout describe the same event with different precision, and both are point-in-time snapshots that markets often partially reverse. Third, narratives about a lab's "decline" tend to overshoot. DeepMind still employs a deep bench and controls Google's enormous compute and distribution. One bad week of headlines is not the same as losing the AI race.

What is genuinely confirmed across all three sources is the core fact pattern: multiple senior, highly credentialed DeepMind researchers left for Anthropic and OpenAI within days of each other, and investors treated it as materially bad news.

The takeaway

The AI race is increasingly being fought in HR, not just in data centers. Compute can be bought and chips can be fabricated, but the people who know how to turn raw scale into a working frontier model are scarce, and they are voting with their feet. For Google, the immediate damage is reputational and financial; the longer-term risk is cultural, if "bureaucratic and risk-averse" hardens into a reputation that deters the next generation of hires. For Anthropic and OpenAI, the windfall is real but unproven — star researchers do not guarantee star results, and integrating Nobel-caliber egos is its own challenge. Watch two things from here: whether DeepMind's next model release closes the benchmark gap its critics describe, and whether this week's exits stay a trickle or become a flood.