Google DeepMind Loses Two Star Researchers in 48 Hours

What happened

• Noam Shazeer, Google's VP of Engineering and co-lead of the Gemini model family, announced on 18 June 2026 that he was leaving for OpenAI. Shazeer co-authored the 2017 "Attention Is All You Need" paper that introduced the Transformer architecture underpinning virtually every major large language model in production today, including GPT, Gemini and Claude.
• John Jumper, a Google DeepMind vice president and co-creator of AlphaFold, announced on 19 June 2026 — the very next day — that he was leaving to join Anthropic after nearly nine years at the lab. Jumper shares the 2024 Nobel Prize in Chemistry for AlphaFold's breakthrough in predicting protein structures.
• Alphabet's stock fell hard on 22 June, dropping as much as 7.2% intraday — its steepest intraday move since February — before closing down around 5%, according to Bloomberg. Reported estimates of the market value erased that day range from roughly $225bn to $270bn, with $250bn the figure most commonly cited across coverage.
• Neither departure has been tied to a stated reason beyond the researchers' own brief public statements, and neither company has detailed compensation or role specifics for the moves.

Two departures, a day apart

The two moves are connected mainly by timing, not by role. Shazeer runs engineering for Google's flagship consumer-facing model family; Jumper ran a scientific research effort that has little to do with chatbots at all. Investors and press coverage treated them as a single story anyway, because together they raised the same question: can Google DeepMind actually hold on to the people who make it the lab it is?

Shazeer's move is the more surprising of the two mainly because of how recently — and how expensively — Google brought him back. He had previously left Google to co-found Character.AI; in 2024, Google reportedly spent around $2.7bn to license Character.AI's technology and bring Shazeer and part of his team back in-house to co-lead Gemini. His second stint lasted under two years. On social media, Shazeer said only that he was "excited" to be joining OpenAI; OpenAI chief executive Sam Altman responded that recruiting him was "only 10 years in the making."

Jumper's departure carries a different kind of weight. AlphaFold isn't a chatbot feature — it's credited with predicting the structure of more than 200 million proteins and reshaping how structural biology and drug discovery are done. Jumper posted on social media that he was leaving "after nearly nine years," crediting DeepMind chief executive Demis Hassabis for "taking a real chance" letting him lead the AlphaFold effort early in his career. His specific role at Anthropic has not been disclosed. Both moves were reported first by CNBC and Bloomberg, with the share-price move confirmed in Bloomberg's 22 June report.

Why the market reacted so hard

Losing one senior researcher rarely moves a multi-trillion-dollar company's share price. Losing two — one of them the co-author of the paper that most of the industry's language models are built on, the other a Nobel laureate whose work sits outside the chatbot race entirely — reads differently to investors than routine attrition. The share-price move on 22 June came days after both announcements, once markets had time to digest that this wasn't a single departure but a pattern forming inside a two-day window.

Part of what makes this reaction notable is what it implies about how investors are now pricing individual researchers. Google's own 2024 decision to spend roughly $2.7bn recruiting Shazeer back helped normalise the idea that a handful of named individuals can be worth billions in strategic value to a frontier lab. Once that logic is accepted, losing those same individuals to a direct competitor becomes a legible, price-relevant event rather than routine HR churn — which is a meaningfully different bar than most companies, in any industry, are held to.

Reactions like this remain unusual. Public companies lose senior technical staff constantly without any measurable share-price effect, because the market has no reliable way to price one employee's contribution against a large existing team. What made Alphabet's case different was less a single data point and more the accumulation of legible signals arriving in one place: the size of the sum Google had reportedly paid to bring Shazeer back, the public prominence of Jumper's Nobel Prize and the AlphaFold body of work, and the tight two-day window separating both announcements. None of those signals is decisive on its own; together, they gave investors a story with a beginning, middle and end that was easy to act on — whether or not it says anything reliable about DeepMind's underlying research capacity.

The wider pattern: a global war for AI research talent

Neither move happened in isolation. Frontier labs — OpenAI, Anthropic, Google DeepMind and others — have spent 2026 recruiting aggressively from each other, often for individual researchers rather than teams. Google DeepMind's own history makes this especially visible: the lab's London roots have already produced more than 100 alumni-founded startups, a talent diaspora that predates this week's headlines by years. What's different about Shazeer and Jumper is that they didn't leave to found something new — they moved directly to two of DeepMind's most direct competitors, taking their specific expertise with them in a way that's much easier for outsiders to price and react to than "researcher leaves to start a company."

DeepMind's research output — including projects like AlphaEvolve, its Gemini-powered algorithm-discovery agent — has long depended on a small number of researchers whose work is unusually easy to recognise and attribute individually, which is precisely what makes them portable. That portability cuts both ways: it's what let Google identify and recruit Shazeer back in 2024, and it's what let OpenAI and Anthropic each make a credible, well-publicised case for hiring him and Jumper away less than two years later.

What it means for AI builders in India and the UK

For builders working outside the three labs at the centre of this story — in Bengaluru, Pune, London or Manchester — the immediate financial stakes here don't apply directly. Nobody outside a handful of frontier-lab researchers is fielding nine-figure counteroffers. But the underlying dynamic scales down in a way that's worth taking seriously.

Shazeer and Jumper could move between the world's most prestigious AI labs in a matter of days because their work was already unambiguous and publicly verifiable — a landmark paper with a fixed publication date, a system that predicted 200 million protein structures, a Nobel Prize. Nobody had to take their capability on faith. That's the extreme case of a much more general principle: in a market where AI skills are reportedly among the hardest to hire for anywhere, and where demand for verified, provable expertise is outstripping supply of people who can demonstrate it, a track record that a hiring manager can check in minutes is worth more than one they have to take on trust.

That's a modest, practical takeaway from a story that is mostly about a handful of people at the very top of the industry. It doesn't require reading too much into two individual career decisions to notice that even the best-funded, most talent-dense lab on the planet couldn't guarantee it would keep either researcher — which is a reasonable prompt for anyone earlier in their career to ask whether their own work is documented well enough that someone else could vouch for it without asking them first.

That distinction between demonstrated and claimed capability shows up constantly at a much smaller scale — in interview loops, promotion cases and freelance pitches. Two engineers with similar years of experience can present very differently to a hiring manager if only one of them has a public trail of shipped work, measurable outcomes or peer-reviewed output to point to. None of that requires a Nobel Prize or a landmark paper; it requires making the ordinary evidence of good work findable, rather than buried in private repositories, internal wikis or closed messaging channels.

What to watch next

Google has not announced any leadership or organisational changes to Gemini or the AlphaFold team in response to either departure, and both products continue to operate. The open questions worth tracking are whether further senior researchers follow Shazeer and Jumper out of DeepMind in the coming months, whether Google responds with visible retention moves for remaining research leadership, and what — if anything — Jumper's undisclosed role at Anthropic turns out to involve. None of that is confirmed yet, and this article will not speculate on it further than that.