Google Loses Two More Senior AI Researchers to Anthropic

Two senior Google researchers who helped build the Gemini AI model are leaving for Anthropic, according to a report on the talent flow out of DeepMind, the latest in a string of high-profile AI departures from Alphabet that rattled investors this week. Jonas Adler and Alexander Pritzel, both viewed internally as key contributors to Gemini, are set to join the Claude maker within months. The move is the second in two days; Nobel laureate John Jumper also joined Anthropic, and transformer co-author Noam Shazeer is heading to OpenAI. None of the four named researchers responded to requests for comment, and Anthropic declined to comment.

The timing is no accident. Anthropic recently raised fresh funding at a $965 billion valuation, confidentially filed for an IPO, and is now positioned to offer even well-paid Big Tech researchers a shot at pre-IPO equity. Alphabet shares closed down slightly on Wednesday after falling as much as 1.2% during the trading day.

The Four-Name Pattern

Bloomberg’s June 24 report named Jonas Adler and Alexander Pritzel as the latest to go. Adler worked on Google’s AI coding effort, the software tools that let machines write and review code on their own. Pritzel was involved in the broader work of training AI systems on massive datasets, a foundational step that turns raw text and images into something a model can reason over. Both are expected to join Jumper at Anthropic.

The pattern is now four senior researchers in recent days. Jumper, who shared the 2024 Nobel Prize in Chemistry for the AlphaFold work that predicts protein structures, had become one of the most public faces of Google’s AI ambition. Shazeer, the co-author of the 2017 paper that introduced the transformer architecture, co-led Gemini after returning to Google in 2024.

A fifth exit, public but quieter, was Arthur Conmy. The senior research engineer at Google DeepMind, who contributed to the Gemini 2.5 model and to AI coding work, said on X on Wednesday that he was joining Anthropic to work on AI safety. Adler and Pritzel both worked with Jumper on the protein-folding research and are set to join him at Anthropic. None of the four Bloomberg-named researchers responded to requests for comment, and Anthropic declined to comment.

• Jonas Adler, AI coding effort at Google, headed to Anthropic
• Alexander Pritzel, AI training systems at Google, headed to Anthropic
• John Jumper, Nobel laureate and AlphaFold lead, headed to Anthropic
• Noam Shazeer, Gemini co-lead and transformer co-author, headed to OpenAI

Why Google Lost Shazeer

The most expensive departure of the week is also the most personal. Shazeer left Google in 2021 to found Character.AI, a chatbot startup, then returned in 2024 under an unusual licensing deal that valued his company at $2.5 billion. Google did not buy the company outright; it hired Shazeer and brought much of the Character.AI team in-house, a structure that let the search giant sidestep the breakup questions an outright acquisition would have raised.

The new job lasted less than two years. Two people familiar with the matter told Bloomberg that, shortly before Shazeer announced his move to OpenAI, Google reassigned the computing power dedicated to one of his projects to a London-based DeepMind team. The stated reason was to streamline pre-training work and to boost cross-team collaboration. The practical effect, current and former employees said, was to strip Shazeer’s project of resources at a moment when it was, by one account, achieving promising results on a new AI architecture still based on the transformer.

Shazeer had also become a divisive figure inside Google. Two people said his comments about transgender identity and the Gaza conflict stirred controversy among some colleagues. He had co-authored the 2017 paper that introduced the transformer, a technique that has become a staple of AI development in the years since.

The exit closes a five-year arc. Shazeer co-authored the 2017 transformer paper, left Google in 2021, returned in 2024, and is heading to OpenAI in 2026. The Bloomberg account identifies three pressures in the months before the move: the compute reallocation, the licensing structure, and his standing inside the company. The same week, the company lost three other senior researchers to Anthropic.

Anthropic’s $965 Billion Pull

What changed since Shazeer last left Google is the size of the prize on the other side. Anthropic closed a $65 billion Series H round led by Altimeter Capital, Dragoneer, Greenoaks, and Sequoia Capital, the company said in its funding announcement. The Series H was co-led by Capital Group, Coatue, D1 Capital Partners, GIC, ICONIQ, and XN.

The funding is the easy part. Anthropic confidentially filed for a US IPO on June 1, 2026, with the listing possible as soon as this fall. The company said earlier this month that its run-rate revenue had crossed $47 billion. The S-1 will land as the AI talent market is, by Hassabis’s own description, the most ferociously competitive it has ever been in the tech industry.

There is also the matter of compute. Anthropic has signed agreements with Amazon for up to five gigawatts of new capacity, with Google and Broadcom for five gigawatts of next-generation TPU capacity, and with SpaceX for GPU capacity in Colossus 1 and Colossus 2. The point, for talent, is that a researcher who joins Anthropic today is signing up to a company that has the silicon to actually train what it wants to train.

• Up to 5 GW of new capacity from Amazon
• 5 GW of next-generation TPU capacity from Google and Broadcom
• GPU capacity in Colossus 1 and Colossus 2 from SpaceX

Google competes in search, advertising, Android, cloud, YouTube, browsers, and a dozen other lines of business. Anthropic builds one product, Claude, and is rapidly layering adjacent ones on top of it. The Claude maker, like Google, is also exploring life sciences and healthcare applications, the Bloomberg report said. Senior researchers weighing the two offers will be reading the timing of the IPO and the size of the equity grant against the salary-plus-RSU package on the other side of the table.

The week’s other numbers, from sourced announcements and the Bloomberg report. Alphabet shares fell as much as 1.2% during the trading day on Wednesday before closing down slightly. Anthropic’s run-rate revenue crossed $47 billion earlier in the month, the company said in its Series H announcement.

• $65 billion in Anthropic Series H funding, led by Altimeter, Dragoneer, Greenoaks, and Sequoia
• $965 billion post-money valuation in the Series H
• $47 billion in run-rate revenue, crossed earlier in the month of the Series H
• $15 billion of previously committed investments from hyperscalers, including $5 billion from Amazon
• 1.2% intraday drop in Alphabet shares on Wednesday, before a slight close

The 11-to-1 Drain

The flow of talent is not anecdotal. DeepMind engineers are nearly 11 times more likely to leave for Anthropic than the reverse, according to a 2025 industry analysis from the venture capital firm SignalFire. The same data found engineers from OpenAI were eight times more likely to flow to Anthropic than the other way around. The data is from 2025, which means the 11-to-1 ratio predates the four-name week of departures the Bloomberg report described this week.

The figures describe the underlying flow, and the Bloomberg report describes the specific departures on top of it. The people who leave first tend to recruit the people they leave with. The pipeline is bidirectional on paper; in practice, the data tracks it running almost entirely one way.

The math behind the math is the size of the headcount at the source. DeepMind employs thousands of researchers, and a 1-in-100 attrition event is still a steady pipeline.

The next round of senior names will arrive the same way the last one did, as a string of individual decisions that, in the aggregate, is a flow. The four named departures this week are the visible tip of a labor market the SignalFire data tracks in aggregate. The pipeline does not need a press release to keep running. The data SignalFire published last year describes the same flow that the Bloomberg report describes this week.

Hassabis Pushes Back From Cannes

Google’s counter-narrative came this week from the top. Demis Hassabis, the CEO of Google DeepMind and a co-winner of the 2024 Nobel Prize in Chemistry alongside Jumper, addressed the departures at an event in Cannes.

There’s a lot of talent movement between all the leading labs and we win our fair share of the top talent. We have by far the biggest and broadest research bench of any of the labs out there. It’s a ferociously competitive market right now, the most ferociously competitive it’s ever been in the tech industry.

The argument is the one a market leader is supposed to make: that the headline departures are a small share of a much larger bench, and that the bench itself is the moat. A Google spokesperson pointed reporters to the Hassabis remarks when asked for comment on Adler and Pritzel. The framing leaves the question of which way the bench is moving to a different dataset.

That dataset is the SignalFire 2025 analysis, which puts the ratio at 11-to-1 in Anthropic’s favor. Hassabis’s ‘biggest and broadest research bench’ is a real claim, and so is the data showing it is the source of most of Anthropic’s senior hires.

The Hard Constraints

Not every exit moves at the same speed. Jumper is unlikely to start at Anthropic until next year, the Bloomberg report said, citing a person familiar with the matter. The reason: AI researchers in the UK, where DeepMind’s leadership is based, are subject to lengthy non-compete agreements that are enforceable under British law. Adler and Pritzel, who also worked with Jumper on the protein-folding research, will face similar constraints if they are based in London.

Other members of the protein-folding team have already moved, just not to Anthropic. Some have shifted to Isomorphic Labs, an Alphabet spinout working on AI-designed drugs, the Bloomberg report said. The pipeline is still flowing: Conmy’s X post on Wednesday was a public declaration by a fifth researcher that the move is now routine, and the timing of his announcement lined up with the Bloomberg report.