Google DeepMind sits at second on the ENTRA Talent Index for H1 2026 — behind only Anthropic — with a two-year retention rate of 78%, a total comp ceiling above $1 million for principal research scientists in Mountain View, and 124 open roles as of May 2026 against a global workforce that Revelio Labs estimated at 8,690 employees by Q1 2026. That position is harder to hold than it looks. On June 19, 2026, John Jumper — 2024 Nobel Prize in Chemistry laureate, co-creator of AlphaFold — announced he was leaving DeepMind for Anthropic after nearly nine years. One day earlier, on June 18, Noam Shazeer, Google VP of Engineering and Gemini co-lead, confirmed he was joining OpenAI. The departures, the biggest talent losses DeepMind has absorbed in a single week, arrived in the same news cycle. The research engine that built AlphaFold, Gemini, and the foundational architecture for modern AI is still the most prestigious address in the field. But the address no longer guarantees the talent stays.
Google DeepMind Headcount and H1 2026 Hiring
Google DeepMind's global headcount reached an estimated 8,690 employees by March 31, 2026 (consolidated Google DeepMind entity, inclusive of Google Brain-integrated headcount, per Revelio Labs workforce analytics, Q1 2026) — up from approximately 4,526 at end of 2025 for the legacy DeepMind Technologies Ltd. entity, with the step-change reflecting structural consolidation of Google Brain-adjacent teams under the unified entity throughout 2025. The two-year organic comparison, from 3,209 in 2024, is the cleaner trend line: a near-tripling of the combined org in two years.
The London King's Cross campus at Pancras Square remains the largest single site, with headcount exceeding 2,100 researchers and engineers by May 2026. Mountain View, distributed across Google's main campus rather than a standalone facility, accounts for approximately 1,400 staff. Paris, where DeepMind has maintained a research presence since 2018, has grown to roughly 300, with the office increasingly involved in EU AI Act compliance work alongside frontier research.
Active hiring in H1 2026 concentrates in four areas. Gemini engineering — the infrastructure and post-training teams behind Gemini 2.0 and the Gemini Ultra tier — is the largest single hiring surface, with roles distributed across Mountain View and London. Gemini Robotics is the most visible growth hire: DeepMind's models for robotic control built on Gemini's vision-language-action architecture, training on ALOHA and Bi-arm Franka platforms, are generating a specific need for robotics researchers with embodied AI experience that the field cannot supply at volume. The biology AI track, institutionally anchored by AlphaFold but now extending beyond protein structure into drug target identification and molecular simulation, has been partially transferred to Isomorphic Labs — the Alphabet subsidiary spun out of DeepMind with $600 million to commercialize AlphaFold technology — but DeepMind itself retains a structural biology and computational biophysics research group that is still hiring. Safety research is the fourth vector, with open roles specifically titled Gemini Safety at Mountain View and Google's Zurich office.
The companies.ts ENTRA data records 124 open roles as of May 4, 2026, against 560 positions listed on LinkedIn globally. The gap reflects DeepMind's standard practice of posting most roles through Google Careers rather than LinkedIn's interface, which undercounts active searches. The real open role figure is closer to the LinkedIn global number.
What Google DeepMind Pays in 2026
Google DeepMind compensation in London runs on a different instrument than Mountain View: Google RSUs priced in USD on the NASDAQ, paid to London employees in sterling at the prevailing exchange rate, on a four-year vest schedule with front-loading introduced in 2022 (33 percent in year one, 33 percent in year two, 17 percent and 17 percent thereafter). That structure matters because it transfers currency risk to the London employee — a consideration that has gained attention as the pound has traded in the $1.25–$1.29 range through H1 2026.
London base salaries, per Glassdoor and CTAIO data through May 2026, run £130K–£180K for L5 (senior) research scientists and £195K–£260K for L6 (staff) research scientists. Total compensation at L5 London lands at approximately £210K–£290K (~$265K–$365K at current rates), including the RSU component. L6 London total comp reaches £300K–£420K (~$380K–$530K).
Mountain View operates on the same Google level structure with meaningfully higher dollar figures. L5 research scientists in Mountain View earn $475K–$625K total compensation. L6 research scientists land at $650K–$850K. L7 (senior staff) runs $950K–$1.4 million, with base salary of $400K–$475K and the remainder in RSUs. The base salary component at senior levels is not the headline number — it rarely is at frontier labs — but it matters because it determines H-1B visa compliance filings and because it is what relocates to a candidate's monthly paycheck before equity vests.
The London-to-Mountain View gap in USD terms is approximately 30–40 percent at equivalent levels. Google has addressed this with London-specific equity refreshers for retained senior researchers — additional RSU grants outside the standard refresh cycle — but has not closed the structural base differential. The gap is the single most cited retention risk in the London market, per ENTRA sourcing, and it has fueled a persistent transfer pipeline from King's Cross to Mountain View that Google manages internally rather than publicly acknowledging.
The "Googler vs. DeepMinder" compensation distinction that defined the first two years post-merger — where Google Brain engineers on Google's standard SWE ladder earned materially more than DeepMind researchers on the legacy DeepMind pay structure — has substantially converged. DeepMind researchers hired or promoted after January 2024 operate on Google's unified compensation system. The legacy gap affected approximately 400 senior DeepMind employees at the time of merger; most have since been harmonized through promotion cycles or renegotiated offers.
The Retention Thesis: Why Researchers Stay at DeepMind
78 percent. That is the two-year retention rate for Google DeepMind employees hired in the last two years, per SignalFire data — second to Anthropic's 80 percent and well above OpenAI's 67 percent. In a market where a senior AI research scientist can receive a credible offer from five different frontier employers in 30 days, that number reflects something structural about DeepMind's position.
Three factors explain it. The first is mission alignment: DeepMind's research record is unmatched. The lab produced AlphaFold 2 and 3, Gemini, the original DQN that started the modern deep reinforcement learning era, and hundreds of Nature and Science papers that carry genuine academic weight. For researchers who built their careers on the academic publication ladder, DeepMind remains the institution where serious science still happens inside a commercial lab — a quality that Anthropic is now competing for but that DeepMind has held for a decade.
The second is Hassabis. Demis Hassabis, in a statement responding to John Jumper's departure announcement on June 19, wrote: "What we achieved with AlphaFold changed the world, and showed the field what was possible with AI for science and medicine, lighting the way for how AI can benefit humanity." The statement is characteristic: it claims the territory of shared achievement rather than institutional grievance. Hassabis, who told Axios in May 2026 that humanity is "standing in the foothills of the singularity," runs DeepMind with the scientific credibility of a Nobel-adjacent researcher and the strategic positioning of someone who has been building toward AGI longer than most of his current competitors have existed as companies. That combination — founder-scientist-strategist — is a retention mechanism that cannot be replicated by comp bands alone.
The third is Google's structural stability. DeepMind has been "largely shielded" from the broader Google workforce reductions that eliminated 12,000+ roles company-wide in 2026, per multiple reports tracking the cuts. For a researcher evaluating counterparty risk across labs, Alphabet's $112-billion annual operating income in FY 2024 (Alphabet 10-K, February 2025) is a different kind of certainty than a venture-backed Series G. Anthropic's retention advantage over DeepMind comes from mission intensity and equity upside, not stability — and for the researchers who prioritize long-term institution-building over near-term equity returns, DeepMind's Google affiliation is a feature, not a constraint.
The Jumper and Shazeer departures test that thesis in real time. Both left for reasons that the retention framework partially predicted: Jumper toward the mission intensity of Anthropic's science program, which has acquired a biotech company and built wet lab infrastructure specifically to pursue the AI-for-science agenda that Jumper led at DeepMind. Shazeer toward OpenAI's IPO equity story. Neither cited dissatisfaction with DeepMind's research environment. Both cited what was pulling them forward, not what was pushing them out. The distinction matters for how Hassabis frames the response.
What's Next
Three things to watch in H2 2026:
1. Biology AI leadership succession. Jumper's departure leaves the biology AI track — AlphaFold 3 extensions, molecular dynamics, drug target identification — without its most prominent public figure. Isomorphic Labs absorbs the commercial side, but DeepMind's internal biology research group will need a credible heir. The hire or promotion that fills that gap in H2 2026 will signal whether DeepMind rebuilds the team around Jumper's methodology or reorients toward a different research direction.
2. Gemini engineering talent stability. Shazeer's exit from the Gemini co-lead role creates the same succession question on the model development side. The Gemini 2.0 Ultra release cadence and the Gemini Robotics expansion are both staffed around teams he helped build. DeepMind and Google will not publicly acknowledge organizational disruption; the signal will come from whether Gemini's publication and release pace sustains through Q3 2026 or shows compression.
3. London comp convergence. The 30–40 percent London-to-Mountain View gap has been the unresolved structural problem in DeepMind's retention architecture since the 2023 merger. In H2 2026, with two high-profile senior departures in a single week and a talent market that has fully priced the premium Anthropic and OpenAI pay, the pressure on Google to close that gap — or to formalize the London-specific equity refresh as a permanent retention tool rather than a discretionary one — becomes harder to defer.
Google DeepMind enters H2 2026 as the most accomplished AI research institution in the world by output record, the second-strongest employer by ENTRA retention data, and the organization most visibly testing whether scientific prestige is still the most durable retention mechanism in frontier AI. The answer arrives in the next six months.
Global Career PlatformFind AI talent. Find your next role.
Booking is hotels. · Airbnb is apartments. · ENTRA is global careers.