Stay informed with free updates
Simply sign up to the Artificial intelligence myFT Digest — delivered directly to your inbox.
The UK will invest in a huge expansion of government-owned AI computing capacity over five years including building a new supercomputer as it seeks to establish a globally competitive artificial intelligence sector, ministers will announce on Monday.
The move is in response to a newly published report on AI opportunities for the UK economy, commissioned by the government and drafted by British venture capitalist Matt Clifford.
The supercomputer will join the UK’s two other advanced machines including Isambard-AI at the University of Bristol, which contains about 5,000 graphics processing units (GPUs), specialised chips to build AI software, and Dawn at the University of Cambridge.
Clifford’s report advocates reaching the equivalent of 100,000 GPUs in government-owned capacity by 2030.
The new capacity, which would represent a 20-fold increase in the UK’s sovereign computing power, will be separate to privately owned AI data centres and will be deployed by the government primarily for AI applications in academia and public services.
It is unclear how much the project will cost, although it will come out of the Department for Science, Innovation and Technology’s research and development budget.
The announcement comes as Clifford is appointed as a part-time adviser to ministers on AI, helping to implement the recommendations in his report, according to two people briefed on the plan. Downing Street declined to comment on the proposals.
Prime Minister Sir Keir Starmer said: “Our plan will make Britain the world leader [in AI]. It will give the industry the foundation it needs . . . That means more jobs and investment in the UK, more money in people’s pockets, and transformed public services. That’s the change this government is delivering.”
Starmer became more excited about the value of AI as an engine of economic growth and public sector reform following a private dinner with former Google chief executive Eric Schmidt and DeepMind chief Sir Demis Hassabis the night before the UK’s global investment summit in October, according to people briefed on the matter.
The Clifford report, known as the “AI Opportunities Action Plan”, was submitted to the government in September, but its publication has been beset by delays. Several cabinet ministers met to discuss its contents in December, according to people briefed on the discussions.
It sets out 50 recommendations to create a thriving national AI industry by improving the conditions to build, scale and adopt the novel technology.
Among the accepted recommendations by the government are: the creation of AI “growth zones”, areas of the UK with accelerated access to planning approvals to build out AI infrastructure; and an AI Energy Council, to advise on requirements around energy resources for AI, including nuclear energy.
Tech experts, including Clifford, have argued that sovereign compute capacity is essential to ensure that British AI companies and researchers can become less dependent on AI businesses in other countries.
They argue that the capacity can establish new AI technology and companies that are globally relevant, and that having access to reliable computer power at a reasonable cost is crucial as computing infrastructure is becoming a geopolitical battleground.
Science and technology secretary Peter Kyle came under fire in August for scrapping funding for an £800mn Exascale supercomputer programme at Edinburgh university, a machine that can perform complex scientific computations such as physics simulations, in a move that caught the technology and academic sector off-guard.
Kyle has insisted he did “not cut anything”, since the £800mn promised by the previous government had not been budgeted for.
In the absence of any significant new sovereign compute programmes, the UK’s most powerful computer has been overtaken by rivals, meaning the country no longer has a machine ranked in the global top 50.
Additional reporting by George Hammond in San Francisco
