Google DeepMind spent more than a decade as the default landing spot for the world’s top machine learning researchers. That run is now measurably over. New staffing data shared exclusively with Fortune and published August 27, 2026, shows the lab’s share of research and advanced-engineering hires across Europe, the Middle East and Africa collapsed from 49% in 2022-23 to just 18.6% in 2025-26, according to UK-based data intelligence firm Zeki Data. Globally, DeepMind’s ratio of new arrivals to departures in research roles fell from roughly 12-to-1 in the second quarter of 2023 to about 2-to-1 in the third quarter of 2026.
The numbers land alongside a string of high-profile exits, a leadership shake-up that saw Demis Hassabis step down from the DeepMind CEO role on August 5, 2026, and mounting reports that Google’s flagship AI lab is losing the researchers who built its reputation to Anthropic, OpenAI, and Meta. This is a straight read of what the new data says, why it matters for the wider AI market, and what it signals about where elite research talent is actually flowing in 2026.
Don't miss new tech stories on Google
Add Tech Insider once in the Google app and our stories appear in your news suggestions.
What the Zeki Data numbers actually show
Zeki Data tracks the flow of engineering and research talent across major AI labs by analyzing hiring and departure patterns. Its founder, Tom Hurd, told Fortune that the drop in DeepMind’s European hiring share is “the sharpest market-share drop that Zeki recorded for a major AI lab in any region.” Hurd’s exact framing of DeepMind’s fall from grace: “They had the crown in Europe forever, and then it started to erode from a very high base.” He added that the erosion is not a single-competitor story: “The likes of Microsoft AI Superintelligence and Meta Superintelligence are eating into their market share, and then there’s OpenAI and Anthropic on the side.”
The headline figure, a drop from 49% to 18.6% of EMEA research and advanced-engineering hires over three years, is a market-share number, not a raw headcount collapse. DeepMind is still, on net, adding more researchers globally than it loses. But the trajectory of that ratio, from roughly 12 new hires for every departure in Q2 2023 to about 2-to-1 by Q3 2026, is the kind of number that keeps recruiting leads up at night. It means DeepMind now needs six times as many open roles to offset the same rate of attrition it absorbed easily three years ago.
Zeki’s data also breaks down where departing DeepMind researchers land. Of people who left DeepMind in the past 12 months, 25% went to Anthropic, 21% to Meta, and 14% to OpenAI, according to the figures shared with Fortune. Anthropic’s own arrivals-to-departures ratio for comparable research and advanced-engineering roles sits at 22-to-1, a full order of magnitude ahead of DeepMind’s current 2-to-1.
Growth rates tell a starker story than headcount alone
Raw headcount numbers can mask how fast a company is actually growing relative to its rivals. On that measure, the gap is wide. Annual research and engineering headcount growth, per the Zeki figures reported by Fortune, breaks down as Anthropic at 152%, OpenAI at 97%, and Google DeepMind at 27%. In an industry where compute access and headline model quality are increasingly gated by how many senior researchers a lab can field, a growth rate five times lower than your closest rival is a structural disadvantage, not a rounding error.
There’s a more specific wrinkle inside the numbers too. Among researchers leaving DeepMind, 19.2% specialized in large language models, compared with 15.6% of those joining. That gap suggests DeepMind isn’t just losing people, it’s losing a disproportionate share of exactly the specialists it needs most as the industry’s center of gravity shifts toward frontier LLM development and away from the broader research portfolio DeepMind built its reputation on.
The AlphaFold exodus: 13 of 29 named authors gone
Nowhere is the departure pattern more visible than around AlphaFold2, DeepMind’s protein-structure-prediction breakthrough and the project that helped earn the lab a Nobel Prize. According to the Fortune reporting, 13 of the 29 named authors on the original AlphaFold2 paper have left DeepMind since publication. That includes John Jumper, the AlphaFold co-inventor and Nobel laureate, who joined Anthropic. Two other AlphaFold veterans, Jonas Adler and Alexander Pritzel, followed him out the door in the same stretch.
The AlphaFold team was, for years, DeepMind’s proof of concept, evidence that fundamental research and product-agnostic science could still thrive inside a company under intense commercial pressure. Losing nearly half its named authors is a different kind of signal than losing generalist engineers: it suggests the erosion is hitting the parts of DeepMind that made it distinct from a standard product lab in the first place.
Who’s actually left, by name
The Fortune investigation, drawn from accounts of current and former DeepMind staffers plus the Zeki data, names a run of senior departures that stretches well beyond AlphaFold. Jeff Dean, a 27-year Google veteran and the company’s chief scientist, is reported to have been leaving alongside “several other prominent researchers,” a detail that broke in the same news cycle as the DeepMind CEO transition. Sanjay Ghemawat, a Google senior fellow, is also named among the departures. Oriol Vinyals and Quoc Le, two long-tenured DeepMind researchers, reportedly launched a venture called Discovery Loop. David Silver, best known for leading the AlphaGo program, founded a new startup, Ineffable Intelligence, which Fortune’s reporting values at $5.1 billion.
Separately, Fortune’s August 13 report on DeepMind’s “identity crisis” notes that in a single week in June 2026, Google lost Gemini co-lead Noam Shazeer to OpenAI, in addition to Jumper’s move to Anthropic. Indian Television Dot Com reported on August 13 that the two original technical co-leads of Gemini also left the company to start a new venture of their own, a detail that lines up with the broader pattern of senior Gemini and DeepMind staff striking out independently rather than moving to a single rival.
Why researchers say they’re leaving
Fortune’s reporting, based on accounts from three current and two former DeepMind staffers, points to a handful of overlapping causes rather than one single trigger. Rival labs, particularly Meta and Microsoft AI, have been offering aggressive, cash-heavy compensation packages. There’s also frustration described inside Google over the company’s competitive position in the AI race, alongside what sources characterized as sinking morale. And there’s the pull of pre-IPO equity: both OpenAI and Anthropic remain private, which means early researchers who join now are betting on a payout structure that looks different from Alphabet’s public-market compensation.
One unnamed former DeepMind engineer, quoted in Fortune’s identity-crisis piece, framed the shift in priorities bluntly: “Most old timers who joined DeepMind before the ChatGPT moment…suddenly were asked to build products for Google.” That captures a tension that has been building inside DeepMind since its 2023 merger with Google Brain: researchers who joined for open-ended science increasingly found themselves reassigned to Gemini product work. Fortune also reports that DeepMind tightened its publication policies, including a six-month embargo on sensitive papers, a move that research-first hires tend to view as a red flag rather than a compliance formality.
Leadership churn compounds the talent problem
The staffing data didn’t emerge in a vacuum. On August 5, 2026, Demis Hassabis stepped down from the Google DeepMind CEO role in what Fortune described as a major AI leadership shake-up. Fortune’s follow-up reporting on August 10 traced the unraveling to stalled models, missed deadlines, and staff burnout inside the lab. For an organization whose recruiting pitch has always leaned heavily on scientific prestige and a stable, research-first culture, a CEO transition landing in the same month as a wave of senior exits is the kind of compounding signal that’s hard for a talent brand to absorb quietly.
This isn’t entirely new. Storyboard18 reported back on July 23, 2025, that Mustafa Suleyman’s move to lead Microsoft AI “marked the start of a slow exodus of DeepMind talent,” with Suleyman bringing several former colleagues along to build out Microsoft’s own AI strategy. What the August 2026 Zeki data adds is scale and a timeline: this wasn’t a one-off poaching story, it’s an accelerating, multi-year trend that has now shown up clearly in the aggregate hiring numbers.
How the four major labs compare on talent flow
| Lab | Annual R&E headcount growth | Arrivals-to-departures ratio | Notable recent signal |
|---|---|---|---|
| Anthropic | 152% | 22-to-1 | Top destination for DeepMind departures (25%) |
| OpenAI | 97% | Not disclosed | Hired Gemini co-lead Noam Shazeer, June 2026 |
| Meta (Superintelligence Labs) | Not disclosed | Not disclosed | Second-largest destination for DeepMind departures (21%) |
| Google DeepMind | 27% | 2-to-1 (down from 12-to-1 in Q2 2023) | CEO transition, Aug 5, 2026; 13 of 29 AlphaFold2 authors gone |
Source: Zeki Data figures as reported by Fortune, August 27, 2026.
DeepMind’s European hiring share, three years apart
| Metric | 2022-23 | 2025-26 | Change |
|---|---|---|---|
| Share of EMEA research/advanced-engineering hires | 49% | 18.6% | -30.4 points |
| Global arrivals-to-departures ratio | ~12-to-1 (Q2 2023) | ~2-to-1 (Q3 2026) | -83% ratio compression |
| Share of leavers specialized in LLMs | Not tracked separately | 19.2% of leavers vs. 15.6% of joiners | +3.6 point deficit |
| AlphaFold2 named authors remaining | 29 of 29 | 16 of 29 | -13 departed |
The market impact: why this matters beyond DeepMind
Alphabet is not short on capital or compute. What Zeki’s data captures is something harder to buy back quickly: a shift in where the industry’s most sought-after research talent believes it can do its best work and get compensated for it. When Anthropic’s headcount growth runs nearly six times DeepMind’s, and its arrivals-to-departures ratio sits at 22-to-1 against DeepMind’s 2-to-1, that’s a compounding advantage. More senior researchers joining a smaller organization means each new hire has outsized influence on research direction, publication strategy, and model architecture decisions, exactly the kind of influence that made DeepMind attractive to top talent for over a decade.
For Google, the practical risk isn’t a sudden collapse in Gemini’s competitiveness. It’s a slower erosion: fewer breakthrough papers carrying a DeepMind byline, a widening gap in the specialized LLM talent pool relative to OpenAI and Anthropic, and a recruiting pitch that has to compete not just on compensation but on a research culture story that took a visible hit with Hassabis’s exit and the tightened publication policy. Public reporting on Nvidia’s roughly $12.9 billion move to acquire Hugging Face, also breaking in the same week according to Fortune, underscores how aggressively capital is chasing AI infrastructure and talent right now. DeepMind’s hiring slide is happening at the exact moment competitors are flush with cash to spend on exactly the researchers it’s losing.
Historical context: how DeepMind built its talent dominance
DeepMind’s European hiring dominance wasn’t accidental. Founded in London in 2010 and acquired by Google in 2014, the lab built a reputation over more than a decade as the place where fundamental AI research, not product shipping, was the point. AlphaGo’s 2016 win over Lee Sedol and AlphaFold’s protein-folding breakthrough became recruiting tools in their own right: researchers wanted their names on papers that reshaped entire scientific fields, not just internal product metrics. That reputation is precisely what let DeepMind claim roughly half of all EMEA research hiring in 2022-23, an extraordinary concentration for a single employer in a competitive field.
The 2023 merger of DeepMind and Google Brain into a single unit was framed internally as a way to concentrate Google’s AI resources against OpenAI’s post-ChatGPT momentum. What the current data suggests is that the merger, and the subsequent push to commercialize Gemini, also changed what DeepMind was for in the eyes of the researchers it depended on to keep its recruiting edge. The tension between “build the next scientific breakthrough” and “ship the next Gemini update” appears to be the fault line running through most of the exits reported in 2026.
What Anthropic and OpenAI are doing differently
Anthropic’s 22-to-1 arrivals-to-departures ratio and 152% annual growth rate reflect a company still in aggressive land-grab mode, with the appeal of pre-IPO equity and a research-first brand that, fairly or not, has absorbed some of the culture DeepMind researchers say they’re losing. OpenAI’s 97% growth and its high-profile hire of Gemini co-lead Noam Shazeer in June 2026 point to a similar strategy: use capital and mission framing to pull senior researchers away from a rival that’s visibly distracted by leadership transition and product pressure. Meta’s Superintelligence Labs division, cited by name in Tom Hurd’s comments to Fortune, is the third leg of that competition, absorbing 21% of DeepMind’s departures over the past year even without disclosed headcount growth figures of its own.
None of the three rivals is winning purely on compensation. All three are, to varying degrees, offering something DeepMind researchers say they’ve lost access to internally: room to run fundamental research without a six-month publication embargo, and a growth-stage equity structure instead of a large-cap public company’s compensation ceiling.
Predictions: where this trend heads next
- Expect DeepMind’s EMEA hiring share to keep sliding through early 2027 unless Google makes a visible, public commitment to loosening publication policy or restoring a research-first mandate for a defined portion of the lab.
- Anthropic’s 22-to-1 hiring ratio is unlikely to hold as the company scales past its current headcount; expect it to compress toward single digits by late 2026 or 2027 simply due to base-rate effects, even as absolute hiring stays strong.
- More DeepMind alumni will likely found independent labs rather than joining a single competitor, following the pattern set by David Silver’s Ineffable Intelligence and Oriol Vinyals and Quoc Le’s Discovery Loop, both of which emerged from the same wave of departures.
- Google will likely respond with retention packages targeted specifically at remaining senior research staff, particularly around the LLM specialists identified in Zeki’s data as overrepresented among leavers.
- Watch for a leadership stabilization push at DeepMind in Q4 2026 or Q1 2027, as a new CEO’s first priority will almost certainly be arresting the visible attrition data rather than any single product milestone.
What this means for engineers evaluating AI lab offers
For software engineers and researchers weighing offers across the major AI labs, the Zeki data adds a concrete data point to a decision that’s usually made on compensation and mission alone. A lab’s arrivals-to-departures ratio is a proxy for internal confidence: when senior researchers are leaving faster than a company can replace them with people of comparable seniority, that tends to show up first in team stability and roadmap churn, well before it’s visible in a company’s external product output. DeepMind’s compressed ratio doesn’t mean the lab has stopped shipping competitive research, Gemini remains one of the industry’s leading model families, but it does mean prospective hires should weigh publication policy, team tenure, and research autonomy alongside salary when comparing offers from DeepMind, Anthropic, OpenAI, and Meta’s Superintelligence group.
Frequently Asked Questions
Is Google DeepMind actually losing employees overall, or just market share?
Based on the Zeki Data figures reported by Fortune, DeepMind is still hiring more research and advanced-engineering staff globally than it loses, so headcount is not shrinking outright. What’s declining is its relative share of top hires and the ratio of new arrivals to departures, which fell from about 12-to-1 in Q2 2023 to roughly 2-to-1 in Q3 2026.
Who is David Silver and where did he go after leaving DeepMind?
David Silver led DeepMind’s AlphaGo program. According to Fortune’s reporting, he founded a new startup called Ineffable Intelligence, which the report values at $5.1 billion.
Why did Demis Hassabis step down as Google DeepMind CEO?
Fortune reported on August 5, 2026, that Hassabis stepped down amid a broader AI leadership shake-up at the company. Follow-up Fortune reporting on August 10, 2026, linked the transition to stalled model releases, missed internal deadlines, and staff burnout inside the lab, though Google has not publicly detailed a single cause.
Where are most departing DeepMind researchers going?
Per the Zeki data shared with Fortune, 25% of DeepMind departures over the past 12 months went to Anthropic, 21% to Meta, and 14% to OpenAI. The remainder went to a mix of other companies and independent startups founded by former DeepMind researchers.
How many AlphaFold2 researchers have left DeepMind?
According to Fortune’s reporting, 13 of the 29 named authors on the original AlphaFold2 paper have departed DeepMind since publication, including co-inventor and Nobel laureate John Jumper, who joined Anthropic.
Is this talent shift unique to DeepMind, or is it happening across the AI industry?
Zeki Data’s founder Tom Hurd told Fortune the drop is specific to DeepMind in scale, calling it the sharpest regional market-share decline the firm has tracked for any major AI lab. Meta’s Superintelligence Labs is also described as gaining, while Anthropic and OpenAI are the clearest net winners in the current data.
Does this affect Gemini’s competitiveness as a product?
Not immediately. Gemini remains one of the industry’s leading model families, and DeepMind is still net-positive on headcount. The risk analysts point to is longer-term: a widening gap in specialized LLM research talent relative to Anthropic and OpenAI, which could show up in future model cycles rather than current ones.
Sources: Fortune, via Yahoo Finance, Fortune’s DeepMind identity-crisis report, Cryptonomist, Storyboard18, Indian Television Dot Com, Business Report, and FourWeekMBA.
