Who left Google DeepMind in June 2026?

Two of the most senior and recognizable AI researchers in the world left in the same week. (1) Noam Shazeer — VP of Engineering at Google, co-leader of the Gemini AI models, and co-author of the original 2017 Transformer paper that made the entire modern LLM era possible — announced on June 18, 2026 that he was leaving Google to join OpenAI. This is the second time Shazeer left Google; he founded Character.AI in 2021, was recruited back to Google in 2024 in a ~$2.7B Character.AI talent-and-tech deal, and is now departing again less than two years later. (2) John Jumper — Nobel Prize co-winner (2024 Chemistry) for AlphaFold, VP Engineering Fellow at Google DeepMind, after nearly nine years at the company — announced on June 19, 2026 that he is leaving for Anthropic. Both departures in the same week reset the public narrative about Google DeepMind's talent stability.

Why are they leaving Google?

Neither Shazeer nor Jumper has publicly disclosed specific reasons. The reasonable inferences from reporting and prior pattern: (1) Compensation — OpenAI and Anthropic have been aggressive in talent recruitment with multi-hundred-million-dollar packages for senior researchers; (2) IPO upside — OpenAI is reported to be on the IPO path, which makes equity grants potentially much more valuable in a short window; (3) Mission and autonomy — both labs offer different research environments and different scales of resource per researcher; (4) Frustration with Google's deployment velocity — Gemini 3.5 Pro's June 2026 GA was just delayed to July, which suggests internal coordination challenges; (5) The Gemini co-lead structure may have created friction. Bottom line: it's a mix of money, mission, and momentum, with money probably the largest single factor.

Is Google DeepMind really losing the AI talent war?

Demis Hassabis (Google DeepMind CEO) publicly responded on June 23, 2026 via Semafor that 'we're still winning AI talent,' citing DeepMind's large research bench (~4,000 researchers), Google's compute access, and ongoing recruiting wins. The mechanical case for Hassabis: two departures out of 4,000+ researchers don't materially change the organization. The narrative case against Hassabis: when the two departing researchers are Shazeer (Transformer co-author, Gemini co-lead) and Jumper (Nobel laureate), public perception shifts regardless of org-chart math. Reality is somewhere in between — Google DeepMind remains a dominant AI research organization, but the perception advantage it had in 2024-2025 has compressed, and that affects future recruiting, partner confidence, and internal morale.

What does this mean for Google's AI products?

Three practical implications. (1) Gemini 3.5 Pro is delayed from June to July 2026 — the timing is almost certainly coincidental (model releases are planned months ahead) but the optics are unfortunate. The Pro model remains expected with 2M context window and Deep Think reasoning. (2) Future Gemini generations — Shazeer's departure as a Gemini co-lead will affect long-term roadmap coordination; Hassabis and the remaining Gemini leadership will need to fill that gap. (3) Google's AI brand — the recruitment narrative matters for attracting the next generation of researchers; Google needs visible counter-recruiting wins in the next 3-6 months to reset the story. None of this affects current Gemini API usage or Google Cloud AI products in the near term — those are operationally separate from the talent story.

Quick Answer

Why Is Google DeepMind Losing AI Talent? (June 2026)

Published: June 25, 2026

Why Is Google DeepMind Losing AI Talent? (June 2026)

In one week (June 18-19, 2026), Google DeepMind lost two of the most senior and recognizable AI researchers in the world: Noam Shazeer (Transformer co-author, Gemini co-lead) departed for OpenAI, and Nobel laureate John Jumper (AlphaFold co-creator) departed for Anthropic. Demis Hassabis publicly responded on June 23 that DeepMind is “still winning” the AI talent war. The reality is more complex — and matters for Google’s AI products, recruiting, and competitive position over the next 12-24 months.

Last verified: June 25, 2026.

TL;DR

June 18, 2026: Noam Shazeer (VP Engineering, Gemini co-lead, 2017 Transformer co-author) announces departure for OpenAI
June 19, 2026: John Jumper (VP Engineering Fellow, 2024 Nobel laureate for AlphaFold) announces departure for Anthropic
June 23, 2026: Demis Hassabis (Semafor interview): “We’re still winning AI talent”
June 24-25, 2026: Gemini 3.5 Pro general availability slips from June to July 2026
The pattern: Senior, public, mission-defining researchers leaving — not a mass exodus but a perception reset
Why it matters: Recruiting, partner confidence, internal morale, narrative momentum
Why it doesn’t matter (much): Google DeepMind still has ~4,000 researchers; current Gemini products are operationally fine

The departures, in context

Noam Shazeer — the most consequential departure

Shazeer is one of the most important figures in modern AI:

2017: Co-author of “Attention Is All You Need” (the Transformer paper) — the architecture that made every modern LLM possible
2000-2021: At Google (the first time)
2021: Left Google to co-found Character.AI
2024: Returned to Google in the ~$2.7B Character.AI talent-and-tech deal; became VP Engineering and Gemini co-leader
June 18, 2026: Announces departure for OpenAI, less than two years after returning

The fact that Google spent ~$2.7B to bring Shazeer back in 2024 and is now losing him to a direct competitor in 2026 is the single most expensive AI-talent reversal on record.

John Jumper — the symbolic departure

Jumper is the most decorated scientist in Google DeepMind:

2017-2026: At DeepMind, nearly nine years
2020: Co-led the AlphaFold breakthrough on protein structure prediction
2024: Awarded Nobel Prize in Chemistry alongside Demis Hassabis
June 19, 2026: Announces departure for Anthropic

Jumper’s departure is symbolically devastating because he won the Nobel Prize with Hassabis (Google DeepMind’s CEO) — and is now joining a direct competitor. Anthropic is in a high-stakes legal and regulatory battle with the U.S. government, which makes the move costlier still.

Why both researchers left

Neither Shazeer nor Jumper has publicly disclosed specific reasons. From reporting and pattern recognition:

1. Compensation packages

OpenAI and Anthropic have been aggressive in talent recruitment over 2025-2026 with multi-hundred-million-dollar packages for senior researchers. Specific figures aren’t public, but Business Insider, TechCrunch, and Bloomberg have reported individual offers in the $100M-$500M range for the most senior frontier-model researchers.

For Shazeer specifically, OpenAI’s reported IPO path (which would convert equity into liquidity in a short window) likely made the package even more compelling.

2. Mission alignment

OpenAI and Anthropic both offer narrower mission focus than Google’s broader product portfolio. For researchers who want maximum focus on frontier model progress (rather than infrastructure, search, ads, and platform considerations), the smaller labs are more attractive.

For Jumper specifically, Anthropic’s safety-focused mission may align better with his AlphaFold-style “AI for science” research interests than Google’s more product-oriented Gemini roadmap.

3. Resources per researcher

Compute access matters. At Google DeepMind, even with TPU access, you’re competing for resources with thousands of other researchers and product teams. At OpenAI or Anthropic, a senior researcher commands a much larger share of total lab compute.

4. Velocity and autonomy

Both Shazeer and Jumper are at career stages where they want to ship and move fast. Google DeepMind’s coordination complexity (Gemini co-leads, multi-product dependencies, Google’s deployment review processes) is real. Smaller labs can move faster.

5. The Gemini 3.5 Pro delay

The delay of Gemini 3.5 Pro from June to July 2026 (announced this week) suggests internal coordination challenges. The model-release schedule is planned months ahead, so the delay is mechanically independent of the Shazeer/Jumper departures — but the same coordination friction that caused the delay may have contributed to their decisions to leave.

Hassabis’s response

In a June 23, 2026 Semafor interview, Demis Hassabis pushed back:

“We’re still winning AI talent… we have the deepest research bench in the world.”

The factual case for Hassabis:

Google DeepMind has approximately 4,000 researchers
Two departures, however senior, don’t materially change the organization
DeepMind has continued making major recruiting wins through 2025-2026 (most names confidential)
Google’s compute access (TPU pods, Colossus integration via Google Cloud, custom silicon) remains unmatched in some categories

The narrative case against Hassabis:

When the departing researchers are the Transformer co-author and a Nobel laureate, perception matters regardless of org-chart math
Public departures are typically the tip of an iceberg; less-senior researchers leave more quietly
Hassabis’s own public push-back (“still winning”) is itself a signal that the perception had shifted enough to warrant a CEO-level response

Reality is in between. Google DeepMind remains a dominant AI research organization. But the perception advantage it had in 2024-2025 has compressed, and that affects future recruiting, partner confidence, and internal morale.

The broader talent war pattern

The Shazeer and Jumper departures are part of a larger pattern of senior AI talent mobility in 2025-2026:

Researcher	Move	Year
Andrej Karpathy	Tesla → OpenAI → independent	2017 / 2022 / 2024
Noam Shazeer	Google → Character.AI → Google → OpenAI	2021 / 2024 / 2026
Various Meta hires	OpenAI/Google → Meta Superintelligence Lab	2025-2026 (4 OpenAI researchers in June 2026 alone)
John Jumper	Google DeepMind → Anthropic	2026

Meta has been particularly aggressive — building its Superintelligence Lab with multi-hundred-million-dollar packages for senior researchers throughout 2025-2026. OpenAI has been recruiting around its IPO path. Anthropic has been recruiting around its safety-focused mission and rapid revenue growth (~$30B run-rate as of May 2026).

The pattern: senior researcher mobility is high, multi-million-dollar packages are normalized, and no lab — including Google DeepMind — has talent stability locked in.

What changes for Google’s AI products

Near-term (next 6 months)

Gemini 3.5 Flash continues to be available and works fine
Gemini 3.5 Pro GA slips to July 2026 (likely independent of these departures but optically connected)
Gemini API and Vertex AI customers see no change — service quality, pricing, support all unchanged
Google Cloud AI products continue normal release cadence

Medium-term (6-18 months)

Gemini 4.0 roadmap — Shazeer was a co-lead; the gap matters here
Recruiting velocity — Google needs visible counter-wins in the next 3-6 months to reset the narrative
Partner confidence — enterprise customers will ask harder questions about Google’s long-term AI commitment

Long-term (18+ months)

Cultural and organizational adjustments — the loss of two of DeepMind’s most-public researchers will reshape how the organization presents itself externally
Open-source vs closed strategy — Google has been more closed than open with frontier models; the talent-war dynamic may push toward more visible research outputs to attract recruits

What this means for everyone else

For OpenAI

Shazeer’s hire is a clear win — both for the engineering impact and the talent-recruiting signal. Expect OpenAI to continue aggressive recruitment through its IPO window.

For Anthropic

Jumper’s hire is a credibility win on the “AI for science” front. Anthropic has been building out non-frontier-LLM research; Jumper fits that direction.

For Meta

The Superintelligence Lab strategy continues. Expect more multi-hundred-million-dollar hires through 2026-2027.

For non-frontier customers

Nothing changes operationally. Continue choosing models based on capability, pricing, and ecosystem fit. The talent narrative affects long-term roadmap confidence but not next-quarter usage.

Bottom line

Google DeepMind isn’t collapsing — it remains one of the largest and most capable AI research organizations in the world. But the perception advantage it built in 2024-2025 has compressed in mid-2026. Two of its most public, mission-defining researchers left in the same week. The Gemini 3.5 Pro delay added optical pressure. Demis Hassabis’s “we’re still winning” response acknowledges the perception shift.

For Google to reverse the narrative, watch for three things in the next 3-6 months: (1) Gemini 3.5 Pro shipping with strong benchmarks in July, (2) visible counter-recruiting wins announced publicly, (3) Gemini 4.0 progress that re-establishes Google as the frontier-model leader.

If Google can do all three, the “talent war” narrative resets. If not, the perception drift continues — and that has real long-term consequences for recruiting, partner confidence, and brand position in the AI ecosystem.