UKRI's £39 Million Turing AI Fellows: Why a Dark-Matter Physicist and an Economic Historian Won AI Fellowships
November 18, 2026 · 6 min read
Granted Research Team · Editorial policy
The fellowship call stated its eligibility rule plainly: applicants must be established researchers "without a background in core AI research." Anyone who had "spent most of their career working on development of frontier AI models" was excluded.
On November 16, 2026, UK Research and Innovation announced the winners. The group includes a dark-matter physicist, a turbulence modeller, a behavioural ecologist, and an economic historian building infrastructure for archival data. In all, UKRI named 20 Turing AI Fellows backed by "more than £39 million in government funding." Seventeen hold the new Turing AI Pioneer Interdisciplinary Fellowships, which started in October 2026 and run three years. Three hold Turing AI Global Fellowships, which begin in 2027, run five years, and bring senior AI researchers to the UK from abroad.
It would be easy to file this as another British AI headline. The design is more interesting than that. UKRI is paying established scientists from outside AI to become AI researchers, and it supplies each one with a mandatory AI co-lead and a large compute allocation. The call also funded more fellows than it said it would. Research offices in the UK, and US agencies weighing similar models, should take notes.
A Call That Funded More Fellows Than It Promised
The Pioneer fellowships were run by EPSRC under UKRI's IS8 AI Priority Programme. The full-application call, which was invite-only after an outline stage, opened December 16, 2025 and closed February 24, 2026. It set aside £22 million for "up to 12" fellowships. Individual awards ranged from £500,000 to £2,187,500 at full economic cost, with UKRI paying the standard 80%.
UKRI funded seventeen, about 40% more fellows than the call promised. The release gives one combined figure for both schemes, so it is impossible to say exactly how much the Pioneer budget grew. The direction, though, is clear. In December 2025, Times Higher Education reported UKRI's spending plans: AI-designated funding rises from £143 million in 2026-27 to £397 million in 2029-30, inside a strategic-priorities budget that grows 13% while curiosity-driven research stays "broadly flat." Fellowship calls funded from a budget that nearly triples over four years can grow after they close, as this one did.
The individual awards are large. CASRAI, citing Imperial College London, reports that Nan Li's materials-forming fellowship is worth £1.8 million over three years, with 18 industry partners committing about £1.2 million in kind. Each fellow also receives an allocation from a cohort-wide pool of 6 million GPU hours on the UK's AI Research Resource (AIRR), which works out to 500,000 hours per fellow. For a physicist or ecologist who has never trained a large model, that compute is arguably worth as much as the salary.
Who Won, and What the List Reveals
The 17 Pioneer fellows sort into five groups:
- Physical sciences and engineering (7): neutrino precision measurement (Thiago Sogo Bezerra, King's College London), AI-run quantum sensing experiments (Cristian Bonato, Heriot-Watt), subsurface characterisation for energy infrastructure (Roisin Buckley, Glasgow), functional materials dynamics (Katherine Inzani, Nottingham), manufacturable-by-design materials forming (Nan Li, Imperial), rare-event detection drawn from dark-matter searches (Jocelyn Monroe, Oxford), and turbulence computation (Jacob Page, Edinburgh).
- Climate and hazards (3): flood forecasting (Qiuhua Liang, Loughborough), near-real-time global greenhouse gas evaluation (Matthew Rigby, Bristol), and gravity-wave inference for climate models (Corwin Wright, Bath).
- Life sciences and medicine (3): cancer digital twins (Jasmin Fisher, UCL), personalised biochemical maps (Michael Inouye, Cambridge), and physics-informed bio-sensing (Jonathan Phillips, Exeter).
- Ecology (2): behavioural ecology for conservation (Sally Keith, Lancaster) and biodiversity forecasting (Andrea Manica, Cambridge).
- Social science and humanities (2): AI infrastructure for historical archives (Neil Cummins, LSE) and bias-aware human-machine collaboration (Ganna Pogrebna, Queen's University Belfast).
Three patterns stand out. First, domain data is the common thread. Almost every winning project starts with a dataset or instrument that only the fellow's field has. Buckley, a senior lecturer at Glasgow's James Watt School of Engineering, told her university that her team, "supported by ScottishPower Renewables," has assembled "the world's largest curated database of onshore and offshore subsurface data." Reviewers appear to have rewarded people who own a hard scientific problem and the data behind it, not people who propose to apply off-the-shelf models.
Second, seniority was interpreted broadly. The call said that "diverse career paths mean that no standardised eligibility criteria will be set," and the cohort includes both senior lecturers and full professors.
Third, funding went well beyond the usual elite institutions. The 20 fellowships went to 16 institutions. Cambridge, Oxford, and the London colleges hold nine. Loughborough, Lancaster, Exeter, Heriot-Watt, and Queen's Belfast each have one. Scotland has four in total.
The Global Fellows Are Recruited From American Institutions
The three Global Fellows are a recruitment tool, and two of them are moving from the United States. Jonathan May is leaving the University of Southern California for Edinburgh to work on "from language to intent." Denny Wu is moving from NYU and the Flatiron Institute to Oxford to build "a predictive mathematical foundation for scalable deep learning." Roi Reichart is moving from the Technion to Cambridge to work on socially intelligent AI.
Five-year packages that move mid-career AI researchers across the Atlantic fit a broader UK push that includes the £60 million SOFAIR and BOLD fundamental AI labs launched in June. We described the strategy behind it in our analysis of the UK's AI research spending. For US researchers who face tighter NIH and NSF paylines, a five-year Global Fellowship is now a realistic option to consider, not a curiosity.
How to Position for the Next Round
UKRI has not announced a new Pioneer or Global call. Still, the AI budget is set to keep rising through 2029-30, and the IS8 programme has made two flagship announcements since June. Researchers who want to be ready should work on four things.
1. Secure your internal slot early. The Pioneer outline stage used demand management. The University of Bath, for example, was limited to four applications and ran its internal expression-of-interest process from August 1 to September 2, 2025, about six weeks before UKRI's October 14 outline deadline. The real first deadline is your university's internal one. Ask your research office now how it would select candidates, and be ready with a one-page outline and CV.
2. Line up your AI co-lead before the call opens. The call made a named AI collaborator mandatory, and an eligible UK-based collaborator becomes the project's single co-lead. Buckley's co-lead is Professor Mark Girolami. Pairing a domain owner with a senior computational partner is evidently what reviewers wanted. Finding that person in the six weeks before an internal deadline is hard, so start the conversation early.
3. Show you can use the compute. With 500,000 GPU hours per fellow and a 10,000-hour familiarisation allocation, reviewers will expect a credible compute plan from people who have not run one before. Use any small AIRR or institutional allocation you can get now and document the results. A pilot dataset already run through a model is much stronger than a promise to learn.
4. Lead with the problem, and present AI as the method. The winning titles all name a scientific outcome first: flood forecasts, greenhouse-gas inventories, cancer digital twins. Eligibility excludes core AI researchers, so a proposal that reads like a machine-learning methods paper works against the call's premise. The strongest case is a field-defining question your discipline cannot answer without AI, supported by data only you have.
The cohort must also take part in Turing Institute community activities, which the call notes are "subject to renewal." Applicants should budget time for that, not just for research. As EPSRC Executive Chair Charlotte Deane put it, the fellowships give researchers "the freedom and support to push the boundaries of AI itself."
If you are a domain scientist preparing to bring AI into your next fellowship or grant proposal, Granted can help you shape the problem-first narrative that this kind of call rewards.