Sovereign AI’s £100m scheme makes Whitehall an early customer

On 31 August the government opened the first four competitions under its £100m Sovereign AI R&D Procurement Scheme, which makes the state an early customer for British AI startups. Upfront payments are available “where appropriate”, and the companies keep the intellectual property (IP) they create. The chancellor, John Healey, announced the scheme at the G20 meeting of finance ministers and central bank governors in North Carolina.

The four challenges cover productivity in the National Health Service, with the Department of Health and Social Care; the efficiency of AI computing infrastructure, led by the Department for Business, Innovation, Science and Trade and the Advanced Research and Invention Agency’s Scaling Inference Lab; connecting data and AI across defense systems, with the Ministry of Defence; and agent security and resilience testing, with the National Cyber Security Centre (NCSC).

The sum is small next to what the public sector already spends on AI. The question for founders, and for the leaders buying AI in public bodies, is whether a pilot-stage scheme can shift a market whose biggest contracts still go to large, established suppliers.

What the scheme offers

The government’s case is that size, not quality, keeps smaller firms out. Its announcement says they are too often locked out of national and local government contracts because they “lack the turnover, cash reserves or track record of larger firms”. The scheme is meant to remove those barriers, with upfront payment as the main lever, and to let successful companies take what they build beyond the pilot and sell it in the UK and abroad.

It targets demonstrator-stage technology: solutions that have moved past early research and are ready to be tested in real operating environments before wider deployment. Applications will be assessed by the Sovereign AI team, the participating departments and independent technical experts against published criteria. The government describes the first competitions as a test of the delivery model, with further challenges to follow.

The scheme sits inside Sovereign AI, the unit the government runs like a venture fund. By 3 August it had made equity investments in five startups, the latest being London-based AI chip designer OLIX, and had backed 11 in total once help with access to compute is counted. “When we said that Sovereign AI would put the heft of a nation behind Britain’s AI founders, we meant it,” said Kanishka Narayan, minister for artificial intelligence, on 31 August.

Where public AI money goes now

Independent data shows the scale of the market the scheme is trying to open. Public bodies awarded 2,129 AI-related contracts worth £5bn between 1 January 2018 and 3 August 2026, according to Tussell’s AI Procurement Tracker, which draws on the procurement data company’s own platform. This year is already the largest on record, with £1.41bn across 453 contracts by 3 August against £1.18bn for the whole of 2025.

The money is concentrated. Microsoft, with £1.03bn of AI-related contract value since 2018, and Palantir, with £628m, are the largest suppliers in Tussell’s data, followed by KPMG, Hochtief and the Laboratory of the Government Chemist. Tussell’s own gloss on that list is “large multiservice providers”. AI also remains a sliver of public technology buying, at roughly 4% of the £123bn of IT services and software contracts awarded between January 2018 and January 2026.

Tussell finds UK-based suppliers have won just under £2.19bn of AI-related contracts since 2018, against £2.28bn for US-based firms, a gap that was close to £1bn in June. But it credits part of the narrowing to a single £456m Cabinet Office contract with KPMG and EY to set up a National School of Government and Public Service, covering civil service training that includes digital and AI skills. A training contract with two professional services firms says little about whether British AI developers are winning deployments.

Why smaller firms struggle to win contracts

The supply side is there. The government’s Artificial Intelligence sector study 2024, prepared by Perspective Economics and published in September 2025, counted 5,862 AI companies in 2024, up 58% on 2023, and found more than 90% of the firms added that year were SMEs.

The same study recorded the barriers. Interviews for the study “noted significant challenges for smaller AI businesses when competing for public sector contracts”, citing procurement processes that favored larger firms, difficulty demonstrating the experience or scale that government contracts require, and subcontracting arrangements that could put smaller, innovative businesses at a disadvantage. The researchers concluded that such firms often struggle to secure government contracts directly.

The wider procurement rules are moving in parallel. On 5 August the Cabinet Office said the weighting for local social value would rise from 10% to 20% on contracts worth £5m or more, and that the rules had been simplified and the threshold raised to contracts above £1m so that they do not typically apply to the work smaller firms bid for. The government puts the market at £90bn a year.

What founders and buyers should watch

For founders, the terms matter more than the headline sum. Upfront payment addresses the cash-flow problem that makes a public pilot risky for a small company, and keeping the IP means a successful demonstrator can become a product rather than a one-off service. The announcement does not say how the £100m will be split across the competitions or when the first contracts will be awarded, so firms should read the published criteria before committing bid time.

For organizations buying AI outside government, the agent security challenge is the one to follow. The competition, run with the NCSC, is meant to support technologies that help organizations understand, manage and mitigate the risks of increasingly capable AI agents. If it produces tested tools from British suppliers, private-sector buyers may find them on shortlists within a year or two.

For departments, the test comes after the pilots. The scheme makes them the first customer; it does not oblige them to become the second. Without a route from demonstrator to wider purchase, the risk is a set of well-funded proofs of concept that leave Tussell’s supplier rankings unchanged.

The verdict: founders should treat the scheme as a way to win a credible public reference customer, not as a revenue line. The government, for its part, should be judged on how many of these demonstrators it goes on to buy at scale.