Google DeepMind’s AI Talent Lead Has Shrunk in EMEA—from 49% to 18.6%

Google DeepMind has not stopped attracting elite AI talent—but its dominance has clearly weakened. Zeki Data estimates that DeepMind’s share of research and advanced-engineering hires across Europe, the Middle East and Africa dropped from 49% in 2022–23 to 18.6% in 2025–26. That 30.4-point decline marked the sharpest regional fall recorded for a major AI laboratory in Zeki’s analysis. 2

The result is a shift in the competitive landscape: DeepMind remains a major research organization, but Anthropic and OpenAI are capturing more of the growth and more of the attention in frontier AI hiring.

DeepMind is still hiring—but losing momentum

The regional hiring share is only one part of the picture. Globally, DeepMind continues to hire more research and advanced-engineering staff than it loses. However, its arrivals-to-departures ratio reportedly fell from about 12:1 in the second quarter of 2023 to roughly 2:1 in the third quarter of 2026. 23

Its growth rate also lagged behind its fastest-moving rivals. Zeki-linked reporting put DeepMind’s hiring growth at about 27%, compared with 97% for OpenAI and 152% for Anthropic. 1

That distinction matters. The figures do not show a lab in free fall; they show a still-large employer whose relative advantage is eroding as rivals expand faster.

Retention is not DeepMind’s biggest weakness

A separate two-year retention comparison presents a more nuanced picture. Anthropic led with 80%, followed closely by DeepMind at 78%. OpenAI recorded 67%, while Meta stood at 64%. 14

DeepMind’s retention rate is therefore not dramatically worse than Anthropic’s. The more serious problem is the combination of slower hiring growth, a much narrower arrivals-to-departures ratio and an unfavorable flow of employees toward competitors.

Zeki’s analysis found that engineers were almost 11 times more likely to move from DeepMind to Anthropic than in the opposite direction. 14 Among DeepMind’s recent departing researchers and advanced engineers, reporting based on Zeki’s data identified Anthropic as the destination for 25%, followed by Meta at 21% and OpenAI at 14%. 412

Which prominent people have left?

Several high-profile exits have made the broader talent trend difficult to dismiss:

• John Jumper, a senior DeepMind researcher and co-creator of AlphaFold, announced that he was leaving for Anthropic after nearly nine years at Google DeepMind. Jumper shared the 2024 Nobel Prize in Chemistry with Demis Hassabis for work connected to AlphaFold.
• Noam Shazeer, a Gemini engineering leader and co-author of the influential 2017 transformer paper, left Google for OpenAI.
• Jeff Dean, Google’s longtime chief scientist, departed alongside Sanjay Ghemawat, Quoc Le and Oriol Vinyals. The group left to found Discovery Loop, an independent public-benefit AI company focused on applying AI to scientific and engineering research. 1719

These departures are significant not simply because of the individuals involved, but because they span DeepMind’s scientific, infrastructure and Gemini efforts.

Why are researchers moving?

The strongest evidence points to an increasingly aggressive market for AI talent. Rival labs are offering substantial compensation, including valuable startup-style equity, while Anthropic has become a particularly important destination for researchers leaving both DeepMind and OpenAI. 1415

The available reporting also places the exits alongside a broader organizational question: how much autonomy should DeepMind retain as Google prioritizes Gemini products, model deployment and commercial execution? Google’s AI reorganization has brought research, frontier models, the Gemini app and developer teams into a more tightly coordinated structure. 17

That structure could help Google ship and commercialize AI more effectively. But the supplied evidence does not prove that publication restrictions, Gemini-related pressure or corporate integration caused any particular individual to leave. Those factors are best treated as organizational context—not quantified explanations for the exodus.

Hassabis is stepping back from daily control

In August 2026, Demis Hassabis moved away from DeepMind’s principal day-to-day management role. He became chair of Google DeepMind and Alphabet’s chief scientist, while Koray Kavukcuoglu took operational responsibility for the AI unit. Kavukcuoglu reports directly to Google CEO Sundar Pichai and oversees Gemini development, frontier research, the Gemini app and developer teams. 1719

The change formalizes a closer connection between DeepMind’s research agenda and Google’s broader product organization. It also gives Hassabis more room to focus on longer-term scientific and AI goals, including his continuing work with Isomorphic Labs. 19

DeepMind still leads in AI for science—but faces pressure in general models

DeepMind’s scientific record remains a major advantage. AlphaFold’s protein-structure work is one of the lab’s clearest contributions, and the significance of Jumper’s move to Anthropic is partly that it involves a central figure from this area.

The lab’s position is less secure in broadly deployed frontier models and coding. A benchmark assessment cited by Fortune placed Google’s best shipped Gemini model behind leading offerings from Anthropic, OpenAI, xAI, Meta and at least one Chinese laboratory. That is a time-sensitive comparison rather than a permanent ranking, but it shows why Gemini execution has become central to Google DeepMind’s competitive position.

What the numbers mean for the AI race

The clearest conclusion is not that DeepMind is collapsing. It is that elite AI talent is now mobile, and DeepMind no longer owns the market it once dominated—especially in EMEA.

Google still has major advantages in infrastructure, distribution and scientific depth. But those advantages must now compete with Anthropic’s retention and recruiting momentum, OpenAI’s rapid expansion and the appeal of smaller frontier labs. 114

DeepMind’s next challenge is cultural as much as technical: prove that tighter integration with Google makes researchers more effective without removing the autonomy and scientific identity that helped make the lab attractive in the first place.