Skip to main content
ScruticaAn open reference on AI compute

Sovereign AI programme of United Kingdom

United Kingdom sovereign AI programme

The United Kingdom has one system of 5,448 accelerators in operation.

AI Opportunities Action Plan, 2025–2030; also covers UK Compute Roadmap, AIRR, Sovereign AI Unit and AI Hardware Plan

  • operating
A full bar is 30,000 accelerators
StageRun byAcceleratorsCountChip
In operationUniversity of Bristol5,448NVIDIA GH200
In operationUniversity of Cambridge1,024Intel Max 1550

Funding

The government’s budget documents state an allocation of £2 billion, 2026 to 2030.

Chips and export rules

Accelerators
Intel’s accelerators are in one of its two systems, and NVIDIA’s in one.Chip makers’ product pages
Fabrication
TSMC fabricates the Hopper parts that one of its systems names.Chip makers’ product pages
US export rules
The United Kingdom is in , outside the groups where exporting advanced AI chips needs a licence from the US Bureau of Industry and Security (BIS).Electronic Code of Federal Regulations, 29 Oct 2026
Scrutica’s rating
Scrutica rates the programme’s dependence on NVIDIA hardware and on US technology high, on a scale of none, low, medium and high.A qualitative assessment of the programme’s chips, partners and supplier relationships, as none, low, medium or high

Partners

12 partners

5 partners by procurement stage, besides the operators in Systems, and 7 more named in the programme’s sources.

Selected, 1
StackHPC
Named in sources
Nscale, CoreWeave, Microsoft, British Business Bank, EPCC, Balderton Capital (James Wise, Sovereign AI Unit chair), AI Safety Institute

Sources read to 8 Sep 2026.

How the figures were made

Operating compute / delivered hardware. Reported operational or delivered: 5,448 NVIDIA GH200 (physical hardware). The stage is assigned from reported hardware status, disbursements and commitments.

Isambard-AI’s full 5,448-GH200 system opened in Bristol on 17 July 2025, providing research compute through the AI Research Resource alongside Cambridge’s Dawn. The July 2025 Compute Roadmap targets 420 AI ExaFLOPS of public capacity by 2030, up from 21 in 2025. Sovereign AI launched in April 2026 with a £500M fund, direct equity investment and access to public supercomputers for startups. The June 2026 hardware plan includes a mixed-chip AI machine targeted for 2030; Edinburgh’s planned £750M national supercomputing service is a separate project.

A full bar is 30,000 accelerators, the scale of Sovereign AI Programmes. Each system is a bar of its own and no total is printed, since two records can describe one machine.

Caveats, 4

  • The July 2025 Compute Roadmap commits up to £2B for public research compute through 2030, including over £1B for AIRR expansion and up to £750M for the Edinburgh service. The 2025 AIRR plan budgeted approximately £250M for AI compute infrastructure and scheduled procurement competitions.
  • The £1.1B June 2026 hardware plan includes £750M for a new AI machine, with £400M of that for chips; £120M for hardware innovation; £45M for skills; and up to £150M for a British Business Bank investment fund. The plan is additional to Edinburgh’s £750M service; its overlap with other parts of the £2B Compute Roadmap budget is unspecified.
  • The Roadmap, published on 17 July 2025, reported £44B of private investment in AI data centres over the preceding 12 months.
  • The installed Isambard-AI count is the full system reported by Bristol in July 2025. The cited Edinburgh announcements do not specify an accelerator count.

Evidence, 10 entries

Hardware records

NVIDIA GH200

United Kingdom

Vendor: NVIDIA

Count: 5,448

Status: Operational

Scope: Physical hardware

Organisation and role

University of Bristol (Isambard-AI full system, launched July 17 2025)

Source

Whole Isambard-AI system. Bristol and NVIDIA describe the same machines as host and chip supplier. Primary source, 17 Jul 2025

Funding

Government spending

No paid figure stated in the government’s documents

The Spending Review 2025 allocates £2 billion for AI over 2026–27 to 2029–30, but no outturn against it has been published; the component line for the AI Research Resource’s first phase carries UKRI’s reported AIRR spending of £64 million (2023–24) and £137 million (2024–25).

Spending Review 2025 allocation for artificial intelligence, 2026-27 to 2029-30 (AI Opportunities Action Plan)

A part of the programme. The Treasury’s allocation that funds the AI Opportunities Action Plan, including at least a 20-fold expansion of the AI Research Resource and the Sovereign AI Unit (SR2025 paras 4.26–4.27), is the budget instrument closest to the record. It is a component because the record also names the Edinburgh national supercomputer, which the Spending Review lists separately (up to £750M), the £1.1bn AI Hardware Plan of June 2026, and the first AIRR phase (Isambard-AI and Dawn, separate line). The same document also says the settlement provides ‘over £2 billion’ for the Action Plan (para 5.109). The period began on 1 April 2026; no outturn for 2026-27 had been published on 2 November 2026. No Estimate line isolates this allocation: DSIT’s 2025-26 accounts report outturn only by broad Estimate line (e.g. ‘Science and Research’).

In a budget estimate: the £2 billion Spending Review 2025 allocation for AI over 2026–27 to 2029–30, April 2026 to March 2030. Printed as “£2 billion”. About $2.7B at the reference rate of 11 Jun 2025. 4.26 At this SR, the government is allocating £2 billion for artificial intellgence (AI) from 2026-27 to 2029-30, fully delivering on the government’s commitment to harness the transformative potential of AI, as set out in the AI Opportunities Action Plan. HM Treasury, HM Treasury, Spending Review 2025 (CP 1285, June 2025), print version, Chapter 4, para 4.26, p. 34 (PDF page 41), 11 Jun 2025

Announced: £2 billion, 11 Jun 2025. Printed as “over £2 billion”. The same Spending Review states the Action Plan funding as ‘over £2 billion over the SR period’ in DSIT’s settlement (para 5.109); kept because the document states the quantity two ways. 5.109 This settlement provides over £2 billion over the SR period to implement the AI Opportunities Action Plan in full, including: HM Treasury, HM Treasury, Spending Review 2025 (CP 1285, June 2025), print version, Chapter 5, DSIT settlement, para 5.109, p. 88 (PDF page 95), 11 Jun 2025

Announced: £500 million, 11 Jun 2025. Printed as “Up to £500 million”. Sovereign AI Unit funding inside the Action Plan settlement; footnote 8 (p. 87) says the fund is £500 million in total, of which £410 million falls between 2026-27 and 2029-30. An allocation, not spending; no amount invested by the unit has been published in any document read. b. Up to £500 million for the creation of a new UK Sovereign AI Unit working with the British Business Bank to support the emergence of national AI champions; HM Treasury, HM Treasury, Spending Review 2025 (CP 1285, June 2025), print version, Chapter 5, para 5.109(b), p. 88 (PDF page 95), 11 Jun 2025

Announced: £140 million, 28 Oct 2026. Printed as “£140 million”. Budget 2026 adds £140 million for the Sovereign AI Unit (£60 million for the Sovereign AI Fund from 2027-28 and £80 million for its R&D procurement scheme, paras 6.99 and 6.105). An allocation, not spending. It will invest an additional £140 million in the Sovereign AI Unit to back promising UK AI companies to scale and succeed here. HM Treasury, HM Treasury, Budget 2026 (October 2026), print version, Executive summary, p. 6 (PDF page 10), 28 Oct 2026

Announced: £2 billion, 17 Jul 2025. Printed as “up to £2 billion”. The Compute Roadmap’s public-compute envelope to 2030: a different composition from the Spending Review’s £2 billion for AI, since it includes the Edinburgh service (up to £750 million) that the Spending Review lists separately. Not reconciled in any document read. Invest up to £2 billion to deliver a diverse, joined-up and user centred compute ecosystem – including over £1 billion to expand the AI Research Resource (AIRR) 20x by 2030 and up to £750 million for a new national supercomputer service in Edinburgh. Department for Science, Innovation and Technology, Department for Science, Innovation and Technology, UK Compute Roadmap (July 2025), web-optimised PDF, ‘The Compute Roadmap at a glance’, action 1, p. 5 (PDF page 6), 17 Jul 2025

Announced: £1.3 billion, 23 Oct 2026. Printed as “£1.3 billion”. Ministerial written answer giving the AIRR expansion commitment as £1.3 billion, against ‘over £1 billion’ in the July 2025 Compute Roadmap; a commitment, not spending. The answer gives no period. The Government has committed £1.3 billion to expand the AI Research Resource (AIRR), providing researchers and innovators with free access to world-class AI compute. UK Parliament (written questions and answers), House of Commons written question 32918 (tabled 9 October 2026), answer of 23 October 2026, Department for Business, Innovation, Science and Trade, Answer, first sentence (record retrieved from the Parliament written-questions API; the web page refused automated access with HTTP 403)

Announced: £750 million, 11 Jun 2025. Printed as “Up to £750 million”. The next national supercomputer at Edinburgh, listed in the Spending Review separately from the £2 billion for AI; an allocation, not spending. DSIT’s 2025-26 accounts report a £21.3 million capital underspend ‘in AI Group due to the supercomputer project’ (p. 50) but no outturn for the project. d. Up to £750 million for a new supercomputer at Edinburgh University. HM Treasury, HM Treasury, Spending Review 2025 (CP 1285, June 2025), print version, Chapter 4, para 4.24(d), p. 34 (PDF page 41), 11 Jun 2025

Announced: £1.1 billion, 8 Jun 2026. Printed as “£1.1 billion”. The June 2026 AI Hardware Plan, including £750 million for a new national AI supercomputer to be deployed in 2030 (of which £400 million for chips); the release says it is additional to the £750 million Edinburgh service. Its relation to the Spending Review allocation is not stated. Not spending. Ambitious £1.1 billion AI Hardware Plan to back British firms developing the chips and computing power behind AI Department for Science, Innovation and Technology, DSIT press release, ‘A decisive shift to power British AI: new £1.1 billion plan to back chip firms, boost computing power and skills for the AI revolution’, 8 June 2026, Summary bullets at the top of the release

AI Research Resource, first phase (Isambard-AI and Dawn), delivered by UKRI from DSIT Managed Programmes funding

A part of the programme. The first AIRR phase built the two systems the record describes (Isambard-AI at Bristol, Dawn at Cambridge). DSIT announced the envelope as £300 million on 1 November 2023 (a written answer of 15 March 2024 repeats it, gb/dl-10); no Estimate line or act stands behind it and no document states its period, so it is an announced envelope with no period, and UKRI’s annual AIRR outturn (2023-24 and 2024-25, printed in its 2025-26 accounts) is kept as figures rather than set against it; UKRI reports a further, unquantified payment in 2025-26 for the ‘completion of AIRR phase one’, so the two years understate the phase’s total. UKRI’s 2024-25 accounts gave the 2024-25 figure as £131 million; its 2025-26 accounts give £137 million; both are kept.

Announced by the government: the £300 million AI Research Resource investment announced in November 2023. Printed as “£300 million”. About $364M at the reference rate of 1 Nov 2023. The investment into the AI Research Resource has been tripled to £300 million, up from £100 million announced in March 2023, in a bid to further boost UK AI capabilities. Department for Science, Innovation and Technology, DSIT press release, ‘Technology Secretary announces investment boost making British AI supercomputing 30 times more powerful’, 1 November 2023, Release body, paragraph beginning ‘The investment into the AI Research Resource’

Spent: £64 million, 31 Mar 2024. Printed as “£64 million”. UKRI’s AIRR spend in 2023-24 (part of DSIT Managed Programmes spend); the 2024-25 accounts print the same £64 million. £64 million of the DSIT Managed Programme spend in 2023-24 was in relation to new funding for AI Research Resource (AIRR). The overall DSIT Managed Programmes spend increased in 2024-25, with AIRR spend increasing to £137 million. UK Research and Innovation, UK Research and Innovation, Annual Report and Accounts 2025-26 (web-accessible version), Performance Overviews, p. 54 (PDF page 56), 14 Jul 2026

Spent: £137 million, 31 Mar 2025. Printed as “£137 million”. UKRI’s AIRR spend in 2024-25 as restated in its 2025-26 accounts. £64 million of the DSIT Managed Programme spend in 2023-24 was in relation to new funding for AI Research Resource (AIRR). The overall DSIT Managed Programmes spend increased in 2024-25, with AIRR spend increasing to £137 million. UK Research and Innovation, UK Research and Innovation, Annual Report and Accounts 2025-26 (web-accessible version), Performance Overviews, p. 54 (PDF page 56), 14 Jul 2026

Spent: £131 million, 31 Mar 2025. Printed as “£131 million”. UKRI’s 2024-25 accounts state 2024-25 AIRR spend as £131 million; its 2025-26 accounts state £137 million for the same year. Both are UKRI records; the difference is not explained. £64 million of the DSIT Managed Programme spend in 2023-24 was in relation to new funding for the AI Research Resource (AIRR). The overall DSIT Managed Programmes spend increased in 2024-25, with AIRR spend increasing to £131 million UK Research and Innovation, UK Research and Innovation, Annual Report and Accounts 2024-25, large-print volume 1, Performance Overviews, financial review of Managed Programmes, p. 54 (PDF page 56), 15 Jul 2025

Searched, 6 places

UK Research and Innovation annual reports and accounts (GOV.UK): UKRI Annual Report and Accounts 2025-26 (published 14 July 2026) and 2024-25 (15 July 2025, large-print volume 1). AIRR spend £64 million in 2023-24 and £137 million in 2024-25 (2025-26 report; the 2024-25 report says £131 million for 2024-25); 2025-26 spend described only as the ‘completion of AIRR phase one’, no amount. No Isambard-AI, Edinburgh or AIRR-expansion outturn. UK Research and Innovation annual reports and accounts (GOV.UK), 2 Nov 2026

DSIT annual reports and accounts (GOV.UK), incl. statement of outturn against parliamentary supply and core tables: DSIT Annual Report and Accounts 2025-26 (14 July 2026; SOPS 1.1–1.2, financial review p. 50; core tables ODS) and 2024-25 (9 July 2025). Outturn only by broad Estimate line (no AI, AIRR or compute line); 2025-26 narrative: £1 billion committed to expand AIRR 20-fold, Sovereign AI Unit established with up to £500 million, ‘£21.3 million underspend in AI Group due to the supercomputer project’; no programme outturn. DSIT annual reports and accounts (GOV.UK), incl. statement of outturn against parliamentary supply and core tables, 2 Nov 2026

HM Treasury Spending Review 2025 and Budget 2026: Spending Review 2025 (11 June 2025), paras 4.24–4.27, 5.108–5.109 and footnote 8; Budget 2026 (28 October 2026), executive summary and paras 4.121, 4.141, 6.99–6.106. £2 billion for AI from 2026-27 to 2029-30; up to £500 million Sovereign AI Unit (£410 million in 2026-27 to 2029-30); up to £750 million Edinburgh supercomputer; Budget 2026 adds £140 million for Sovereign AI. Allocations only; no outturn. HM Treasury Spending Review 2025 and Budget 2026, 2 Nov 2026

DSIT AI and compute policy papers (GOV.UK): UK Compute Roadmap (17 July 2025; evidence annex replaced 23 April 2026); AI Opportunities Action Plan: One Year On (29 January 2026); AI Research Resource guidance collection (updated 28 September 2026); releases of 1 November 2023 and 8 June 2026. Envelopes only: up to £2 billion public compute to 2030 (over £1 billion AIRR expansion, up to £750 million Edinburgh); up to £250 million earmarked for AIRR cloud capacity; AIRR £300 million (2023); AI Hardware Plan £1.1 billion (2026). No spend to date. DSIT AI and compute policy papers (GOV.UK), 2 Nov 2026

UK Parliament written questions and answers (official API, questions-statements-api.parliament.uk): Answered questions matching ‘Isambard’, ‘Dawn supercomputer’, ‘compute roadmap’, ‘AIRR’, ‘compute’, ‘supercomputer’, ‘Research Resource’, ‘Sovereign AI’, ‘national supercomputer’ (up to 300 results per term; the API returned HTTP 500 for ‘AI Research Resource’ and ‘exascale’), 2023 to 2 November 2026; full texts of answers 18137, 128826, 114085, 16203, 17501, 32918, HL8670. Commitments only (£300 million AIRR 2023; up to £2 billion public compute; £1.3 billion AIRR expansion; capacity of Isambard-AI and Dawn); no answer states AIRR, Edinburgh or Sovereign AI spending to date. AISI compute spend 2025-26 (£3.34 million, answer 37339) is outside the programme. UK Parliament written questions and answers (official API, questions-statements-api.parliament.uk), 2 Nov 2026

National Audit Office and committee reports: Web search for NAO and Science, Innovation and Technology Committee reports on the AI Research Resource, Isambard-AI and public compute (2023–2026). No NAO or committee report giving AIRR or compute spending located. National Audit Office and committee reports, 2 Nov 2026

Still open

UKRI’s 2025-26 AIRR spend (‘completion of AIRR phase one’) is not printed, so total first-phase spending is unknown.

UKRI’s 2024-25 AIRR spend is £131 million in its 2024-25 accounts and £137 million in its 2025-26 accounts; the restatement is unexplained.

How the Spending Review’s £2 billion for AI, the Compute Roadmap’s up to £2 billion, the £1.3 billion AIRR expansion and the £1.1 billion AI Hardware Plan overlap is not stated in any document read.

Written answers since at least 3 September 2026 come from the Department for Business, Innovation, Science and Trade (BIST), which Budget 2026 also names; where its accounts will report AI outturn for 2026-27 is unverified.

No spending has been published for the AIRR expansion (including the up to £250 million cloud capacity), the Edinburgh service or the Sovereign AI Unit’s investments.

Open Sovereign AI Funding

Partners

University of Bristol

United Kingdom

Accelerators: 5,448

Model: NVIDIA GH200 Grace Hopper Superchip

Stage: Commissioned

The count

Physical hardware. Whole Isambard-AI system. Bristol and NVIDIA describe the same machines as host and chip supplier. Primary source, 17 Jul 2025

Notes

Hosts and operates Isambard-AI at the National Composites Centre, Bristol and Bath Science Park. The full 5,448-GH200 system launched in July 2025.

Source

Primary source, 1 Nov 2023

Open University of Bristol’s record

Hewlett Packard Enterprise

United Kingdom

Model: NVIDIA GH200 Grace Hopper Superchip

Stage: Awarded

Notes

Supplied Isambard-AI’s Cray EX system, Slingshot interconnect and ClusterStor storage.

Qualifications

Programme-level disclosure

Source

Primary source, 1 Nov 2023

Open Hewlett Packard Enterprise’s record

University of Cambridge

United Kingdom

Accelerators: 1,024

Model: Intel Data Center GPU Max 1550

Stage: Commissioned

The count

Physical hardware. Dawn’s 256 servers each contain four Intel Max 1550 GPUs; each GPU has two internal stacks. Primary source

Notes

Hosts Dawn at the Cambridge Open Zettascale Lab as part of the AI Research Resource.

Source

Primary source, 1 Nov 2023

Open University of Cambridge’s record

Intel

United Kingdom

Model: Intel Data Center GPU Max 1550

Stage: Awarded

Notes

Supplies the Intel Data Center GPU Max accelerators used by Dawn at Cambridge.

Qualifications

Programme-level disclosure

Source

Primary source, 1 Nov 2023

Open Intel’s record

Dell Technologies

United Kingdom

Stage: Awarded

Notes

Supplies Dawn’s PowerEdge XE9640 servers with direct liquid cooling.

Qualifications

Programme-level disclosure

Source

Primary source, 1 Nov 2023

Open Dell Technologies’ record

StackHPC

United Kingdom

Stage: Empanelled

Notes

Worked with Bristol and Cambridge on Scientific OpenStack, the cloud service that connects Isambard-AI and Dawn for researcher access.

Qualifications

Programme-level disclosure

Source

Primary source, 1 Nov 2023

Open StackHPC’s record

NVIDIA

United Kingdom

Accelerators: 5,448

Model: NVIDIA GH200 Grace Hopper Superchip

Stage: Awarded

The count

Physical hardware. Whole Isambard-AI system. Bristol and NVIDIA describe the same machines as host and chip supplier. Primary source, 17 Jul 2025

Notes

Supplies the 5,448 GH200 Grace Hopper superchips used by Isambard-AI.

Source

Primary source, 1 Nov 2023

Open NVIDIA’s record

Fabrication

TSMC

Fabricates the parts of these families: Hopper

Hopper The full GH100 GPU that powers the H100 GPU is fabricated using the TSMC 4N process customized for NVIDIA NVIDIA, “NVIDIA Hopper Architecture In-Depth”, 2026-09-25

Open the record

Identifiers

Scrutica ID
sovereign-gb
Country code
GB
Name in the record
United Kingdom: AI Opportunities Action Plan / UK Compute Roadmap / AIRR / Sovereign AI Unit / AI Hardware Plan