Technology and AI
Four competitions open a 100 million pound state procurement scheme for British AI firms
By Staff Writer | 1 September 2026
The Treasury has launched the first procurement competitions under a 100 million pound scheme that buys demonstrator stage artificial intelligence from British companies. Successful firms keep the intellectual property they create, and upfront payments are available where appropriate, which removes the two barriers that usually keep small suppliers out of public contracts.
The scheme is called the Sovereign AI R&D Procurement Scheme, and the launch was announced on 31 August by the Chancellor at the G20 meeting of finance ministers and central bank governors in North Carolina. Four competitions opened with it.
The first is an NHS productivity challenge run with the Department of Health and Social Care, for systems that automate workflows, coordinate care and support decision making across health services. The second, on compute efficiency, is led by the Department for Business, Innovation, Science and Trade with the Scaling Inference Lab at the government's advanced research agency, and looks for technology that makes artificial intelligence computing infrastructure cheaper to run. The third, with the Ministry of Defence, is for connecting data and frontier models securely across defence systems. The fourth, with the National Cyber Security Centre, is for testing the security and resilience of increasingly capable software agents.
When we said that Sovereign AI would put the heft of a nation behind Britain's AI founders, we meant it.
Kanishka Narayan, AI Minister
The two terms that matter to a small supplier
Public procurement usually screens on turnover, cash reserves and track record, which is a rational way for a buyer to manage counterparty risk and a reliable way to exclude any company that has not sold to the state before. The announcement says those barriers are removed for this scheme and that upfront payments are available where appropriate. For a company at demonstrator stage, working capital is the constraint, and being paid on delivery against a public sector payment cycle is often the reason a bid is never made.
The intellectual property term is the sharper one. A supplier that keeps what it builds has an asset at the end of the pilot. A supplier that assigns it has a completed contract and nothing to sell.
Retention of intellectual property allows a company to take the same technology beyond the pilot and sell it elsewhere. The government's stated reason for allowing it is that this is how a demonstrator turns into a business, and that a first public customer is the hardest step to cross. Whether that logic survives contact with departmental commercial teams is the question the first round will answer.
What is being bought, and what is not
The target is stated narrowly: technology that has moved beyond early research and development and is ready to be tested in real operational environments, but is not yet deployed. That is a deliberate exclusion of both ends. Pure research is funded elsewhere and finished products are bought through ordinary procurement. Applications will be assessed by the Sovereign AI team, the participating departments and independent technical experts, against published criteria.
Alongside the launch the Chancellor confirmed a plan to open the AI Economics Institute, a government backed body examining the economic effects of artificial intelligence, to cooperation with G7 countries on information and evidence sharing.
The four competitions are described as a test of the delivery model, with further challenges to follow. That is worth taking at face value. A hundred million pounds spread across four departmental challenges is not a large procurement by the standards of any one of those departments; it is an experiment in how the state buys from companies too small to sell to it.