UK backs Oxford and UCL labs to make AI cheaper to run

TL;DR:

  • Oxford and UCL will host two new government-backed AI research labs sharing up to £60m over six years.
  • The labs aim to build open-source models that run on widely available hardware, reducing reliance on a few large providers.
  • The funding sits within a £1.6bn UKRI AI strategy and doubles an earlier plan from one lab to two.

The government has announced up to £60m for two new research labs, led by Oxford and University College London, tasked with making AI cheaper, more reliable and easier for organisations across the UK to adopt. Announced on what would have been Alan Turing’s 114th birthday, the move doubles an earlier plan — from one lab to two, and from £40m to up to £60m — and comes with access to large-scale computing power worth tens of millions of pounds.

Sovereignty through efficiency, not scale

The two labs take deliberately different angles. UCL’s Science of Fundamental AI Research (SOFAIR) Lab, led by Professor David Barber with Cambridge, Oxford and Edinburgh, will build open-source models that run on widely available hardware — potentially even ordinary consumer computers — to reduce dependence on the handful of firms that dominate model provision. Oxford’s British Open-ended Learning and Discovery (BOLD) Lab, led by Professor Jakob Foerster with UCL and Imperial, will rethink how AI learns more efficiently.

Foerster framed the rationale bluntly: “The UK cannot win the global AI race simply by trying to outspend the largest technology companies on data and compute.” AI Minister Kanishka Narayan said building capability at home was about “reducing reliance on others and securing Britain’s place at the forefront of this technology”. The pitch lands directly on the UK’s running sovereignty debate, which Resultsense has tracked through cities like Leeds positioning itself as Britain’s second AI city.

Looking forward

The labs will fund researchers at every career stage, with £2m earmarked per lab to hire at least ten doctoral students, and will work alongside the Alan Turing Institute and UKRI’s AI hubs. The strategy’s logic — that efficiency and open weights are a more credible route to sovereignty than chasing US compute budgets — is sound, but six years is a long horizon in a field measured in months. For UK businesses, the prize is models cheap enough to run in-house rather than rented from providers that, as recent export-control disputes have shown, can switch access off.

Share this article