The UK's AI industrial policy in 2026: what is funded and what is just announced

Two kinds of number

When ministers describe the UK's AI industrial policy, two very different kinds of number get folded into a single headline. There is money the Treasury has committed, and there is money that private companies say they intend to spend. The Department for Science, Innovation and Technology set both out in its report AI Opportunities Action Plan: One Year On, published on 29 January 2026, which said 38 of the plan's 50 commitments had been met and about 68 billion pounds of AI investment had been pledged since January 2025. According to the department, most of that total is private. The eye-catching tens of billions are pledges, not Treasury spending.

The public money is more modest and easier to count. The report lists 240 million pounds for the AI Security Institute, 250 million to expand cloud capacity for the AI Research Resource, 750 million for a national supercomputer in Edinburgh, and up to 500 million for a new Sovereign AI Unit, which began making its first equity investments in April 2026. Each of the five AI Growth Zones received up to 5 million pounds in targeted funding. These are real commitments, but together they sit in the low billions, not the tens of billions that dominate the announcements.

AI Growth Zones: pledges first, delivery later

The government designated five AI Growth Zones, at Culham, the North East, North Wales, South Wales and Lanarkshire. DSIT says they account for 28.2 billion pounds of investment and more than 15,000 projected jobs. Those figures are commitments from companies such as Microsoft, CoreWeave and Vantage Data Centers, contingent on sites being built and powered. The largest single pledge made under the programme is a plan by Nscale, Nebius and CoreWeave to deploy roughly 120,000 Nvidia Blackwell chips, with investment of up to 11 billion pounds. That is a procurement intention, not installed capacity.

Delivery is the harder part. The first zone, at Culham in Oxfordshire, is planned to start at 100 megawatts and scale towards 500, but reporting through 2025 noted that no development partner had been named and the scheme had not entered planning. Across the zones, independent analysis found no general acceleration in planning activity, and the single biggest constraint is electricity. The government itself accepts that the grid connection queue is heavily oversubscribed, with data-centre requests totalling more than 70 gigawatts against only a few gigawatts of connected capacity. A new AI Growth Zone Delivery Unit was set up in late 2025, and the government has used powers in the Planning and Infrastructure Bill to reserve grid capacity for priority projects, an admission that the ordinary process is too slow for the timetable ministers have set.

Compute: an ambition restored after a cut

The compute story is the clearest case of an announcement outrunning the money, then the money catching partly up. In August 2024 the incoming Labour government shelved about 1.3 billion pounds of AI funding inherited from its predecessor, including 800 million for an exascale supercomputer at Edinburgh and 500 million for the AI Research Resource, saying the sums had not been allocated in existing spending plans. The University of Edinburgh had already spent 31 million pounds on a building to house the machine.

The following year the government reinstated a version of the project. The 2025 Spending Review set aside 750 million pounds for the Edinburgh system, but it is no longer exascale. The director of the university's computing centre, Mark Parsons, told The Register that the exascale moment "has passed." DSIT also committed 250 million pounds to scale the AI Research Resource's cloud capacity and 36 million to add AMD chips to the Dawn supercomputer at Cambridge. The department says national AI compute has risen from about 2 to 21 ExaFLOPs, roughly a tenfold increase, with a target of a twentyfold increase by 2030. The largest system, Isambard-AI in Bristol, built around 5,448 Nvidia Grace Hopper superchips, began opening to researchers after its launch in July 2025. The ambition is real, but the restored figure is smaller and later than the one that was cancelled.

From safety to security

The third pillar is the body that began in 2023 as the AI Safety Institute. In February 2025 the technology secretary, Peter Kyle, renamed it the AI Security Institute at the Munich Security Conference, keeping the acronym AISI but shifting its remit towards crime, fraud, child sexual abuse material, cyber-attacks and the misuse of models to help build chemical or biological weapons. It gained a joint research team with the Home Office.

On money and output the institute is one of the better documented parts of the strategy. The Spending Review gave it 240 million pounds, and DSIT says it employs more than 100 researchers, has tested 30 frontier models and published ten peer-reviewed papers at the NeurIPS 2025 conference. It also runs grant programmes, including an Alignment Project launched in July 2025. What the institute still lacks is statutory power. Its testing of frontier models remains voluntary, agreed with the labs rather than compelled by law, so its influence depends on cooperation it cannot require.

What the record shows

A fair reading is that the UK has a genuine AI industrial policy with real public money behind it, but that the money is smaller and more targeted than the headlines imply. The large numbers are private investment intentions that depend on grid connections and planning consent the state does not yet control. The compute target was cut and then partly restored. The safety body has been funded and refocused but not given legal teeth. None of that makes the strategy a failure. It does mean the honest measure of progress over the next year is delivered megawatts, connected data centres and compute hours used, not the size of the next pledge.