How has Google DeepMind’s position as Europe’s dominant employer of elite AI researchers changed from 2022–23 to 2025–26, including the fall in its share of research and advanced engineering hires across Europe, the Middle East and Africa from 49% to 18.6%, the decline in its arr

Google DeepMind has shifted from Europe’s clear first-choice elite-AI employer to one major competitor among several. It remains a formidable research organization with exceptional infrastructure, but its regional hiring dominance, retention advantage, and identity as an unusually open-ended research lab have materially weakened. 2

• Talent-market reversal: Zeki Data estimates that DeepMind’s share of research and advanced-engineering hiring in Europe, the Middle East and Africa fell from 49% in 2022–23 to 18.6% in 2025–26—the sharpest regional share decline it recorded for a major AI lab. 2 Its arrivals-to-departures ratio reportedly fell from roughly 12:1 in Q2 2023 to about 2:1 in Q3 2026. 5 That still means net inflow, but no longer the overwhelming pull it once had as Anthropic, OpenAI and Meta expanded their European recruiting. OpenAI and Anthropic’s hiring growth was reported at 97% and 152%, respectively, versus 27% for DeepMind. 1

• Visible departures: In June 2026, Gemini contributors Jonas Adler and Alexander Pritzel moved to Anthropic, while Transformer co-author and former Character.AI leader Noam Shazeer left for OpenAI. 24 Later, Alphabet chief scientist Jeff Dean, senior fellow Sanjay Ghemawat, Oriol Vinyals and Quoc Le left to found Discovery Loop. 1 The combined effect is more consequential than raw headcount: these were unusually senior people with influence over foundation models, infrastructure and scientific AI.

• Why people are leaving: Reports attribute the shift partly to competitors’ exceptional cash and equity offers, but compensation is not a complete explanation. 14 DeepMind’s traditional attraction was latitude to pursue long-horizon work and publish influential research; its mandate has increasingly centred on delivering and commercializing Gemini, with less autonomy and reportedly tighter controls around publication. The AlphaFold-team restructuring, alongside a pivot toward LLMs and general scientific automation, is a potent signal of that narrower organizational priority. 26 The causal weight of each factor is not publicly quantified, so it would be overstated to describe the exits as caused by any one policy.

• Hassabis’s new role: In August 2026, Demis Hassabis relinquished day-to-day CEO responsibilities to become DeepMind chair and Alphabet chief scientist; CTO Koray Kavukcuoglu took operational control and reports directly to Sundar Pichai. 12 This gives Hassabis a broader company-wide remit, but it also reduces DeepMind’s institutional independence and separates its public scientific figurehead from daily management. That can improve alignment and speed, yet may deepen the perception that the lab is becoming a product arm of Google rather than a distinct research institution.

• Where it remains strong—and where it has ceded ground: DeepMind remains unusually strong in scientific AI, reinforcement learning, multimodal systems, infrastructure, and research talent density; AlphaFold’s legacy is evidence of a distinctive capability in biology and scientific discovery. 6 Its more vulnerable areas are frontier consumer-facing assistants, rapid product iteration, coding agents and the perceived “raw intelligence” of shipped general models: one cited independent benchmark assessment placed Google’s best shipped model behind Anthropic, OpenAI, xAI, Meta and a Chinese lab. 6 That is a snapshot rather than a definitive ranking, but it captures the competitive perception problem.

• Strategic threat: OpenAI and Anthropic combine strong model brands, product momentum and powerful recruiting narratives; Meta can pay aggressively and deploy models across a huge product ecosystem. Microsoft AI Superintelligence adds a well-funded rival focused explicitly on frontier capability. DeepMind’s advantage in Alphabet’s compute, custom TPU stack, distribution and capital is real, but compute only helps if leadership can retain the people who decide what to train, ship and publish. The organizational reshuffle and talent leakage therefore threaten to turn an infrastructure advantage into a less decisive competitive advantage. 12

The central change is not that DeepMind has become weak; it is that its former European monopoly on prestige, freedom and talent has eroded. Its recovery depends on making Gemini commercially successful without eliminating the scientific autonomy that made the lab a talent magnet in the first place.