A policy analysis from mid-2026
When Sir Keir Starmer stood before cameras on 13 January 2025 and unveiled the AI Opportunities Action Plan, he made a characteristically blunt argument: the UK could either shape the AI revolution or be shaped by it. “We will not be a technology taker,” he said. The plan itself , commissioned from tech entrepreneur Matt Clifford and backed by the government’s acceptance of all 50 of its recommendations , was the most comprehensive statement of AI industrial policy Britain had ever produced. Sixteen months on, with 38 of those 50 commitments formally delivered and a live public progress dashboard tracking every milestone, the plan has moved further and faster than many expected. But the harder work , turning infrastructure into adoption, and adoption into growth , is only now beginning.
This piece examines what the Action Plan actually set out to do, what has genuinely been delivered, where the structural gaps remain, and what the next phase looks like for the UK’s AI ambitions as of May 2026.
What the Plan Actually Said
The AI Opportunities Action Plan was built around a simple but ambitious argument: the UK had world-class AI assets — safety institutions, research talent, financial markets, the NHS, the BBC, the National Archives — and was in danger of squandering them through under-investment, regulatory caution, and institutional inertia. Clifford’s 50 recommendations were organised across three broad pillars: investing in the foundations of AI; driving cross-economy AI adoption; and fostering homegrown AI champions.
The government accepted every single recommendation. The ambition was explicit: the UK should aim to become not just an AI user but an AI maker , building sovereign compute capacity, creating a national data infrastructure, producing domestic AI champions that could compete globally, and positioning the country’s approach to safety and governance as a competitive differentiator rather than a drag on innovation.
What’s Working: The Foundations Are Being Built
Compute: From Near-Zero to Serious
A year ago, the UK’s public AI compute capacity was embarrassingly thin. The Action Plan called for a minimum twenty-fold increase in the AI Research Resource by 2030. The One Year On progress report, published on 29 January 2026, confirmed that the 10x increase had already been achieved — from 2 to 21 ExaFLOPs — with a target of 420 ExaFLOPs by 2030. The Isambard-AI supercomputer launched at the University of Bristol in July 2025. In January 2026, the government confirmed a £36 million investment to increase supercomputer capacity at the University of Cambridge’s DAWN facility sixfold by Spring 2026.
The AI Growth Zones are the physical infrastructure complement to this compute buildout. Five zones : Culham in Oxfordshire, the North East, North Wales, South Wales, and Lanarkshire in Scotland — have been formally designated, with planning and energy access reforms to fast-track data centre construction. The Lanarkshire zone alone is backed by £8.2 billion in private investment and is projected to create more than 3,400 high-skilled jobs. The first pilot at Culham is exploring integration of Small Modular Reactors to power data centres of up to 500MW , a deliberate bet on sustainable, sovereign compute infrastructure rather than dependency on overseas cloud giants.
In parallel, the government backed up to £250 million to scale cloud capacity for the AI Research Resource, and launched a call for information on the development of secure AI computing systems, closing in February 2026, to inform the next generation of sovereign infrastructure strategy.
Investment: The Numbers Are Moving in the Right Direction
The commercial case for UK AI is strengthening. According to the government’s own infographic data, UK AI firms raised £6 billion in venture capital in 2025 , up from £4 billion in 2024 and £2.6 billion in 2023, representing 52% annual growth. Since January 2025, £68 billion in investment has been pledged to UK AI. The country remains, by a meaningful margin, the leading AI market in Europe, home to Google DeepMind, ARM, and Wayve among others.
The April 2026 launch of the Sovereign AI Unit , backed by up to £500 million, chaired by James Wise of Balderton Capital, and delivered by DSIT, represents the operationalisation of Clifford’s most ambitious recommendation: a government-backed function with a mandate to invest in UK AI companies, provide access to compute and data, support talent relocation, and actively cultivate the next generation of globally competitive British AI companies.
Data Infrastructure: The National Data Library Takes Shape
One of the plan’s most distinctive contributions is the National Data Library (NDL), backed with over £100 million at Spending Review 2025. The NDL is intended to make the UK’s extraordinary public sector data assets — from HMRC records to planning data to cultural archives — usable for AI development in a governed, sovereign way. It has published guidelines for making government datasets AI-ready and launched its first open calls for data proposals from researchers and industry.
The most transformative element of this is the Health Data Research Service (HDRS): a joint commitment of up to £600 million from the government and Wellcome Trust, creating a secure single access point to regional and national NHS datasets. With Dr Melanie Ivarsson appointed as CEO in January 2026 and Baroness Nicola Blackwood as Chair, the HDRS is moving from concept to institution. The potential here is significant: NHS data is among the most comprehensive population health datasets in the world, and unlocking it — responsibly — could give UK AI researchers and life sciences companies a structural advantage.
A Creative Content Exchange, announced in June 2025 and now in pilot phase with 12 leading cultural institutions including the National Archives, Natural History Museum, and Royal Armouries, creates a marketplace for licensed access to digitised cultural and creative assets — a pragmatic complement to the wider copyright debate.
AI Safety: Substance, Not Just Symbolism
The AI Safety Institute (AISI) has continued to build genuine international credibility. Its Frontier AI Trends Report, ten peer-reviewed papers presented at NeurIPS 2025, and a landmark study on AI-driven persuasion published in Sciencehave cemented its role as a global reference point for AI safety research. By chairing the International Network for AI Measurement, Evaluation and Science, AISI is shaping how advanced systems are evaluated worldwide , amplifying UK influence well beyond its funding levels.
On the assurance side, the government has established an £11 million AI Assurance Innovation Fund, a Trusted Third-Party AI Assurance Roadmap, and a new Centre for AI Measurement at the National Physical Laboratory — the beginnings of a domestic market for trusted AI assurance services. The first round of the Assurance Innovation Fund opened for applications in Spring 2026.
Public Services: Early Proof Points, Not Yet Scale
The NHS is delivering some of the most concrete early results. AI is now being used to halve treatment times for stroke patients, speed up prostate cancer diagnosis, and — according to government data — one third of chest X-rays in the NHS are now AI-enabled. The government’s AI fraud detection tools have delivered an 80% reduction in time to spot fraud risks, reportedly saving half a billion pounds for public services.
The “Minute” AI scribe — an automated transcription and summarisation tool for government meetings — has been scaled to support 1,000 officials across 22 local authorities. The planning tool “Extract,” which processes planning documents in under two minutes, is expected to be available to all councils by Spring 2026. AI tutoring trials have launched in schools, with tenders and co-creation partnerships for in-school tools announced by March 2026.
Where It’s Stuck: The Persistent Gaps
Despite genuine progress on foundations, three structural problems continue to shadow the Action Plan.
The Copyright Impasse: A Battle Not Yet Won
The most politically charged unresolved issue is the copyright and text-and-data mining (TDM) regime. The Action Plan called for UK TDM rules to be made at least as permissive as the EU’s , a prerequisite, in Clifford’s view, for UK AI developers to train models on the data they need. This recommendation has yet to be actioned. It is the only major infrastructure-adjacent commitment that has not moved.
The reason is a genuine policy collision between the AI sector and the creative industries. On 18 March 2026, the government published its Report on Copyright and Artificial Intelligence, fulfilling a statutory obligation under the Data (Use and Access) Act 2025. The report declined to introduce a commercial TDM exception, with Technology Secretary Liz Kendall stating that “it will take time to get this right.” The House of Lords Communications and Digital Committee, in a detailed and fairly critical report published on 6 March 2026, warned of a “clear and present danger” to UK creative industries and firmly rejected any opt-out TDM model. Until this is resolved, UK AI developers training large models face genuine legal uncertainty — and the risk that the most talent-intensive part of the AI value chain relocates to more permissive jurisdictions.
Adoption Rates: Cautious, Uneven, and Concentrated
The government’s own AI adoption research, published alongside the One Year On report, found that AI uptake across the UK economy remains cautious, uneven, and concentrated in narrow, off-the-shelf uses. A 2025 study conducted with Microsoft found that around 65% of public sector organisations were experimenting with AI — but only about 30% had fully integrated it into their operations. Some 41% of public sector staff feel unprepared or unsupported in using AI, and 37% of leaders say they would use AI more with clearer training and safeguards. The gap between piloting and integration is the defining challenge of 2026.
This is not a funding problem. It is a structural one: the skills, confidence, governance frameworks, procurement capability, and institutional appetite to deploy AI at scale across large public sector organisations. As Professor Alan Brown of the University of Exeter has argued, the government’s own data shows that only one in ten major government technology programmes succeeds — and without root-and-branch institutional reform, AI is at risk of replicating that pattern.
Regulation: Deliberate Restraint or Dangerous Gap?
The UK has made a deliberate choice not to follow the EU AI Act model of cross-sectoral, risk-tiered regulation. Instead, it has maintained a sector-specific, regulator-led approach — the FCA, PRA, CMA, ICO, and MHRA each setting expectations within their domains, with regulators now required to report annually on how they have enabled AI innovation and growth. An AI Growth Lab — a cross-economy regulatory sandbox — has been launched to trial AI in real-world settings where current rules create friction.
This approach has genuine advantages: it is faster to operationalise than new primary legislation, and it avoids the risk of a poorly designed statute locking in the wrong rules at the wrong time. But the Labour Government’s July 2024 commitment to introduce “appropriate legislation” for the most powerful AI models has not materialised, and the regulatory landscape for frontier AI development remains uncertain. As Clifford Chance has observed, regulators will need to demonstrate that they can enforce compliance without stifling innovation — and that has yet to be truly tested.
Skills: A Long-Run Race the UK Cannot Afford to Lose
The upskilling ambition is large: 10 million workers to receive AI training by 2030. The pace of early delivery is encouraging — one million AI courses delivered since June 2025, ahead of schedule. The government joined a new AI Skills Partnership with techUK and industry during London Tech Week 2025, targeting 7.5 million workers with essential AI skills. Visa fee reimbursements for AI researchers and academics have been introduced, and the Global Talent Taskforce’s resourcing was doubled in January 2026, bringing in specialist private-sector headhunting expertise.
But structural questions remain. With some 200,000 people currently studying AI-related higher education programmes, the pipeline is real — but the destination matters as much as the supply. Whether the UK can retain the AI talent it trains, and ensure that upskilling translates into real productivity gains across the economy rather than concentration in a narrow tech elite, is an open question that no single initiative resolves.
What Comes Next: The Delivery Phase
The One Year On report was clear about the transition underway. The founding phase of the Action Plan — designating zones, commissioning infrastructure, launching institutions — is largely done. The UK’s ambition, as stated explicitly, is to become the fastest-adopting AI country in the G7. The 2026 roadmap focuses on enabling deployment, encouraging pro-innovation sandboxes, and accelerating AI in priority sectors.
Several directions are now clearly in motion:
Sector-specific AI as a national priority. The Industrial Strategy, backed by £150 million for six transformative AI and technology programmes, is explicitly targeting priority sectors — including professional and business services — for AI adoption support. The BridgeAI programme, expanded at Autumn Budget 2025, is designed to provide tailored guidance and funding to help thousands of businesses deploy AI before the end of this Parliament.
Public services as the proof point. The government’s stated goal is to take proven AI tools in public services — diagnostics, planning, digital assistants — and scale them nationally so that people everywhere benefit, not just in digitally advanced trusts and authorities. The Delivery Unit for AI Growth Zones is in full operation. The challenge is pace: scaling AI nationally in the NHS, local government, and schools requires not just technology but procurement reform, workforce change, and trust.
Sovereign AI as industrial policy. The Sovereign AI Unit, launched in April 2026, is the most explicitly interventionist element of the plan — a government fund with a mandate to take stakes in UK AI companies, provide compute and data access, and actively cultivate domestic champions. It is backed by up to £500 million and represents a significant departure from the UK’s historically arm’s-length approach to technology industrial policy. Whether it can identify and back the right companies — and avoid the pitfalls of state-backed technology investment — will be watched closely.
The copyright question cannot stay unresolved. The March 2026 report kicked the TDM reform into a further consultation phase, with no commitment to a specific legislative timeline. Industry will be watching this closely: the longer the uncertainty persists, the greater the risk that model training activity relocates to jurisdictions with clearer rules. The government’s “genuine reset moment” on copyright needs to land somewhere concrete.
An Action Plan at Its Hardest Moment
The UK AI Opportunities Action Plan was ambitious when it launched. Measured against its own metrics, its first year has delivered more than its critics expected — 38 of 50 commitments met, compute capacity multiplied tenfold, £68 billion in pledged investment, five AI Growth Zones designated, and a Sovereign AI Unit finally operational.
But the harder yards lie ahead. The foundations are largely built. Now the plan has to prove that foundations translate into adoption, adoption into productivity, and productivity into the kind of broad-based economic benefit that justifies the ambition. The UK’s public sector AI adoption remains stuck in the pilot-to-integration gap. The copyright question threatens to become a structural drag on model development. The regulatory framework, while deliberately light, remains untested by any serious enforcement challenge.
As techUK has put it, 2026 is increasingly being seen as the year where delivery, adoption, and implementation come to the fore. The government’s own target — fastest AI-adopting country in the G7 — is a measurable, competitive benchmark. The infrastructure is now in place to try to meet it. Whether the institutions, the regulatory clarity, and above all the scale of practical deployment can match the ambition is the central question of the next chapter.
Key Links
- AI Opportunities Action Plan (January 2025) — GOV.UK
- AI Opportunities Action Plan: One Year On (January 2026) — GOV.UK
- AI Opportunities Action Plan: Progress Dashboard — GOV.UK
- AI Opportunities Action Plan: Infographic & Key Statistics — GOV.UK
- Report on Copyright and Artificial Intelligence (March 2026) — Fieldfisher analysis
- CMS Law: 2026 Progress Report Analysis — CMS
- techUK: Delivery Must Be the Focus in 2026 — techUK
- Clifford Chance: Unpacking the UK AI Action Plan — Clifford Chance
- Reed Smith: AI Opportunities Action Plan & Regulation — Reed Smith
- Osborne Clarke: UK Regulatory Outlook, February 2026 — Osborne Clarke
- AI Growth Zones: Regional Innovation — Digital Government Network
- UK Public Sector AI: Pilot to Integration Gap — ResultSense
