22APR

EPSRC doubles AI-lab spend to £60m

4 min read

17:16UTC

EPSRC launched two AI research labs on 23 June and doubled the programme behind them to £60m, funding open-source AI that runs on ordinary chips rather than the frontier hardware the rest of government is buying.

← Back to Britain's innovation pipe leaks at both ends Jump to analysis ↓

TechnologyAssessed

Key takeaway

EPSRC is funding the AI that works without the infrastructure DSIT is spending billions to build.

The Engineering and Physical Sciences Research Council (EPSRC), the UK's main public funder of engineering and physical sciences, launched two AI research labs on Tuesday 23 June and doubled the programme behind them to £60m, with AI Minister Kanishka Narayan making the announcement ¹. The SOFAIR Lab at University College London will develop open-source AI that runs on commodity hardware, the ordinary chips anyone can buy, cutting dependence on a handful of model providers. The BOLD Lab at Oxford will work on new learning algorithms and embodied AI, the software that runs robots. Each lab gets around £8m to start and £2m for at least ten doctoral students, with an autumn review deciding further tranches.

SOFAIR's mandate quietly cuts against the rest of government policy. DSIT, the Department for Science, Innovation and Technology, is spending £1.1bn on an AI Hardware Plan built around frontier chips and large model providers, with the US firm Playground Global running its £150m chip vehicle. EPSRC is funding the opposite hedge: AI designed to work without that infrastructure. The european-tech-sovereignty topic owns the frontier-policy analysis; the angle here is that the state is funding both sides of its own bet, and neither plan acknowledges the other.

The same anti-dependency reflex named five British AI-hardware startups at RUSI, the Royal United Services Institute defence think tank , and pushed Oriole Networks' photonic interconnect into deployment through ARIA, the Advanced Research and Invention Agency . SOFAIR carries that logic into the research base, assuming the expensive infrastructure might not arrive. BOLD's robotics pillar is the longer bet, a pipeline for the embodied-AI spinouts Britain does not yet have.

Deep Analysis

In plain English

The Engineering and Physical Sciences Research Council ; EPSRC ; is the UK government body that funds academic research in engineering, mathematics, and physical sciences. On 23 June 2026 it announced two new AI research laboratories and doubled its total spending on this programme from £40m to £60m. The first lab, called SOFAIR, will be based at University College London. Its brief is to build AI systems that run on ordinary, widely available computer chips ; the sort any company or university can buy ; rather than the highly specialised and expensive chips that the most powerful AI systems currently require. The second lab, called BOLD, will be at the University of Oxford and will focus on new ways of teaching AI to learn, and on building AI that can control physical robots. The AI Minister, Kanishka Narayan, announced the launch on 23 June. The timing is notable because the government is simultaneously spending over £1bn on advanced AI chips and infrastructure through a separate programme. SOFAIR is, in effect, a research safety net: if that expensive infrastructure does not deliver, there will be a cheaper alternative being developed in parallel.

Deep Analysis

Root Causes

UK AI capability depends on a handful of US frontier-model providers and the chip ecosystem underpinning them. DSIT's £1.1bn AI Hardware Plan (the european-tech-sovereignty topic owns that analysis) attempts to reduce chip dependency, but both the plan and the model ecosystem remain US-controlled. EPSRC's commodity-hardware mandate is the cheapest available hedge: open-weight models on standard chips require no export licence, no hyperscaler relationship, and no frontier-chip supply chain.

BOLD's embodied-AI mandate addresses a separate structural absence. The UK's robotics industry has no equivalent of Boston Dynamics, Agility, or Figure AI at the commercial stage.

Oxford and Imperial produce internationally competitive robotics research but few commercial spinouts, partly because embodied-AI requires the integration of hardware, software and physical actuation ; a combination that demands more capital than a pure-software AI startup and more lab time than a hardware startup. EPSRC funding the foundational learning algorithms is the upstream step before any of that commercial infrastructure can form.

What could happen next?

Opportunity
SOFAIR's open-source commodity-hardware mandate creates a public-domain alternative to frontier-model provider APIs for UK public sector and SME applications, reducing the cost of AI adoption in government and the NHS if the research produces deployable tools.
Medium term · Reported
Risk
The autumn 2026 review gate creates an eighteen-month horizon for both labs to demonstrate publishable output; labs that fail to produce early results face funding termination before doctoral programmes are complete, which may deter the most capable researchers from joining.
Short term · Reported
Consequence
BOLD's embodied-AI programme seeds a UK robotics spinout pipeline that currently does not exist; the first commercial spinouts are likely five to eight years from launch, placing them in the 2031–2034 window.
Long term · Reported
Meaning
EPSRC funding open-source commodity-hardware AI while DSIT funds frontier chips and large model providers means the UK government is simultaneously backing two mutually contradictory AI infrastructure strategies, with neither body acknowledging the other's hedge.
Medium term · Assessed

Source Landscape

This story draws on neutral-leaning sources

Primary parallel: DARPA's 1990s High Performance Computing and Communications programme funded open-source alternatives to proprietary supercomputer architectures when Cray and SGI dominated the market. The programme produced Linux-cluster HPC as a viable open alternative, which eventually displaced proprietary architectures entirely.

SOFAIR's mandate follows the same logic: fund open-source AI on commodity hardware when frontier providers dominate, on the assumption that the open alternative eventually catches up at a fraction of the cost.

Counter-parallel: The UK's 2010s Quantum Technologies programme offers a cautionary read on doubling publicly-funded labs. EPSRC committed to two national hubs ; quantum computing at Oxford, quantum sensing at Birmingham ; in 2014, then doubled the programme in 2019.

The doubling funded good research but did not accelerate spinout formation; the commercial quantum cluster (OQC, Quantum Motion, Riverlane) emerged from different funding roots. BOLD's embodied-AI lab faces the same risk: doubling research capacity does not automatically create the missing commercial layer.

£60m across two labs over what is likely a five-year programme implies roughly £6m per lab per year ; competitive with a well-funded university research group but modest by frontier-AI standards, where a single GPT-scale training run costs tens of millions of dollars. The £2m doctoral component per lab funds ten or more PhD students whose thesis work becomes the first cohort of spinout founders five to seven years from now.

The autumn 2026 review gate is the significant economic decision point. EPSRC has structured this as a staged funding commitment, consistent with a portfolio management approach that has replaced open-competition grants across UKRI.

Labs that produce publishable results and industry partnerships by autumn 2026 will receive further tranches; those that do not will be restructured or terminated. That contingent structure is more capital-efficient than a fixed multi-year award, but it also means neither lab has a secure funding horizon beyond the review.

Consensus view: The Tony Blair Institute for Global Change's 2025 AI infrastructure report and the Resolution Foundation's 2026 technology chapter both identify frontier-model concentration risk as a structural vulnerability: when sovereign AI capability depends on three or four US hyperscalers and model providers, a pricing change, export restriction, or geopolitical disruption disrupts national AI capacity overnight.

SOFAIR's commodity-hardware mandate addresses exactly that dependency. The IPPR's 2025 paper 'Public AI for Public Benefit' specifically recommended EPSRC-led investment in open-weight models that run on commodity hardware, making this launch a direct implementation of a recommendation the left-of-centre research community has pushed for two years.

Counter-view: Onward's 2026 technology competitiveness paper argues the opposite: frontier-model capability compounds at a rate that commodity-hardware research cannot track. Onward's position is that UK sovereignty in AI requires access to the most capable models, not alternatives to them; commodity-hardware open-source AI produces tools ten to twenty times weaker than frontier systems, which means SOFAIR produces research-grade outputs rather than commercially deployable ones.

Policy Exchange's Miriam Levin argued in the FT in May 2026 that splitting state AI investment between a frontier-infrastructure plan and a commodity-hardware hedge produces two under-capitalised strategies rather than one viable one.

Key tension: Whether SOFAIR and BOLD produce frontier-competitive research output ; publishable in Nature, spinnable into companies ; or academic-grade tools that deepen the UK's research reputation without bridging into commercial AI capability at the scale the Hardware Plan is supposed to address.

First Reported In

Update #9 · Private money rebuilds Britain's seed tier

Tech Funding News· 24 Jun 2026

Read original →

Causes and effects

Caused by

Kendall names UK chip five at RUSI

Kendall's naming of five UK AI hardware startups at RUSI reflects the same anti-dependency instinct that EPSRC's SOFAIR commodity-hardware mandate embodies.

Occurred 28 Apr 2026

Read story →

Oriole moves data with light, not copper

Oriole's ARIA-backed photonic interconnect is an earlier instance of the same state bet on AI that does not require frontier chips that SOFAIR's mandate continues.

Occurred 8 Jun 2026

Read story →

UKRI confirms 58% cut to STFC labs

STFC's discovery-science cuts land upstream of the EPSRC AI programme the state expanded to £60m in June.

Occurred 9 Jul 2026

Read story →

This Event

EPSRC doubles AI-lab spend to £60m

A research council is quietly hedging against the £1.1bn frontier-chip bet that another arm of the same government is making.

Different Perspectives

Institute of Physics

The Institute has long argued STFC's national-laboratory infrastructure, not its grant programmes, is the binding constraint on UK physics output, and warns mothballing capacity like Clara removes capability that cannot be rebuilt on a four-year cycle. It represents the discovery-science community absorbing the reallocation the Bank's equity cheques do not touch.

Helsing

The Munich-headquartered defence-AI firm chose Plymouth over Continental sites for a £350m manufacturing plant building underwater surveillance gliders, alongside its record raise. Its choice of postcode signals confidence in UK manufacturing capacity for defence hardware even as it looks abroad for the capital financing that hardware.

Dragoneer Investment Group, Lightspeed Venture Partners and Iconiq

The three US growth-capital firms backed Helsing's $1.8bn round at an $18bn valuation, more than doubling the mark set only a year earlier, with demand reportedly exceeding the capital on offer. Their money, not a UK sovereign vehicle, is what funds the Plymouth plant, extending a pattern of foreign capital underwriting British defence-hardware manufacturing this cycle.

British Business Bank

The Bank wrote its largest-ever direct life-sciences cheque into Alchemab and added a £6.5bn SME lending guarantee the same week UKRI confirmed the STFC cuts. It is deploying an April mandate change letting it lead venture rounds and invest directly up to £60m per company, treating equity extension rounds and small-business debt as newly within its risk appetite.

Daphni

The Paris seed fund joined Speedinvest and three UK backers in Astral Systems' GBP23m Series A for modular fusion reactors, one of the round's five European co-investors betting on lab-to-market fusion ahead of any working commercial reactor. Unlike CuspAI's all-foreign cap table, this round kept a UK lead investor in Mercia Ventures.

EQT

EQT, appointed by the European Innovation Council to run the EUR5bn Scaleup Europe Fund, entered advanced talks for a further CuspAI stake reported on 3 July, the fund's first pursuit of a UK-founded winner. A closed deal would put EU sovereign capital, not a UK vehicle, on the cap table of a company Britain's own funds passed over.