SuperIntelligenceGuide.co.uk
Plain English. Primary sources. Dated and reviewed.
The UK angle

Could the UK build AGI or superintelligence?

An evidence-led look at what Britain has and lacks (laboratories, research, talent, compute, energy, chips, capital and ownership) for building AGI or superintelligence. Not a prediction.

The terracotta and glass facade of Google's building at 6 Pancras Square, King's Cross
6 Pancras Square, King's Cross: Google's London building and the home of Google DeepMind: founded here, owned in California.Photo: Acabashi, CC BY-SA 4.0, via Wikimedia Commons (cropped)

In short: The United Kingdom has the people and institutions associated with frontier AI and lacks most of the physical and financial plant. On the asset side: Google DeepMind was founded in London and is run from there, with AGI as its published aim; OpenAI and Anthropic have their largest research offices outside the United States in London; Arm designs the processors in most of the world’s phones from Cambridge; and British AI companies took 39 per cent of European venture capital in the first half of 2026. On the other side: the models DeepMind builds are trained on Google hardware that has no published UK location and are owned by a US company; the UK’s biggest public AI computer is roughly one-fortieth the size of the largest US training clusters; Britain had the highest industrial electricity prices of any reporting IEA country in 2024; it designs chips but has no leading-edge fabrication capacity; and our review found no published evidence, as of 23 September 2026, of a UK-headquartered company training a frontier general-purpose model in Britain. This page sets out that evidence. It does not predict whether AGI or superintelligence can be built here, because no one knows whether either can be built anywhere.

First, what the question means

“Could the UK build AGI?” runs together three things that are worth separating, because the evidence on each is different.

  • Companies founded or headquartered in Britain. That is about incorporation, leadership and where the staff sit.
  • Research done in Britain. That is about universities, institutes and the laboratories’ London offices.
  • Compute, energy and capital located in Britain. That is about where the training runs physically happen, who owns the resulting model, and who paid for it.

A country can score highly on the first two and near zero on the third, and the United Kingdom is the clearest example in the world of that pattern. The sections below take each in turn, and the table at the end lines them up.

AGI and superintelligence are used here in the research sense, roughly human-level generality and greatly beyond the best humans in virtually all domains, not in the sense of the US administration’s “Super Intelligence”, which is a name for AI in general. As of 23 September 2026 no UK body had used the US term; the one place a UK minister has used “superintelligence” in a speech, David Lammy at the UN Security Council on 24 September 2025 (“superintelligence is on the horizon”), was descriptive, not a goal.

Laboratories: founded here, owned elsewhere

Google DeepMind was incorporated in London on 23 September 2010, bought by Google in January 2014, and merged with Google Brain in 2023 into a single organisation “led by our CEO Demis Hassabis” from King’s Cross. Its published aim is AGI, which it defines as “AI that’s at least as capable as humans at most cognitive tasks”. Hassabis has said: “We founded DeepMind in London because we knew the UK had the potential and talent to be a global hub for pioneering AI.”

Two facts qualify that. DeepMind is a division of Alphabet, and its models are Alphabet’s property. And Google’s published list of the regions where its TPU hardware (the chips it trains Gemini on) is located includes several US regions, the Netherlands, Japan, Taiwan and Chile, and no UK zone. Google’s Waltham Cross data centre, opened in September 2025, is described by Google as serving “Google Cloud, Workspace, Search and Maps”; Google has not published where DeepMind’s training runs take place. Analysts who count “notable” models by country reflect that: Stanford’s 2025 AI Index lists the UK with no notable models in 2024 (the United States: 40; China: 15), which is only possible if Google DeepMind’s models are counted as American. On other counting rules the UK looks stronger (Epoch AI’s June 2024 tally gave the UK 42 large-scale models cumulatively), so the number depends on whether a London-led team counts as British when the hardware and the owner are American.

OpenAI and Anthropic both opened their first overseas offices in London in 2023. In April 2026 OpenAI called London “our largest international research hub” and took a permanent office at King’s Cross for more than 500 staff from 2027; Anthropic in the same month announced an expansion to accommodate up to 800 staff in the Knowledge Quarter. Microsoft opened a London “AI hub” in April 2024 under Jordan Hoffmann, formerly of DeepMind. These are research offices: neither OpenAI nor Anthropic has publicly identified a frontier-model training cluster in the UK, and all three operations report to US parents.

British-founded companies at scale exist but not at the frontier of general models. Wayve (self-driving, $1.05 billion Series C in May 2024, led by SoftBank with Nvidia and Microsoft), Synthesia (video, $200 million at a $4 billion valuation, January 2026), ElevenLabs (voice, $500 million at $11 billion, February 2026, with London among six listed offices) and Isomorphic Labs (drug discovery, $600 million, March 2025, headquartered in London and also led by Hassabis) are large, London-based and specialised. As of 23 September 2026 we found no UK-headquartered company that has trained a frontier general-purpose model in the UK.

Research: strong on people, unsettled on institutions

The UK ranks fourth in the 2024 Tortoise Global AI Index, behind the United States, China and Singapore, with the same index noting that “the UK’s performance in this particular area [model development] is comparatively poor” and that France “currently outspends the UK by 60 per cent” on generative-AI companies. Stanford’s index counted seven UK papers among the hundred most-cited AI publications of 2023.

The public research base is real: UKRI funds twelve AI Centres for Doctoral Training (£117 million, first cohort 2024) and Turing AI fellowships at Edinburgh, Imperial and Nottingham. The national institute is less settled. The Alan Turing Institute received £100 million of UKRI funding over five years in March 2024. In July 2025 the then Science Secretary, Peter Kyle, wrote to its chair instructing it to “reform itself further to prioritise its defence, national security and sovereign capabilities”; around 50 of its 400-plus staff were put at risk of redundancy, and its chief executive, Jean Innes, resigned in September 2025, as reported at the time. The Institute’s own announcements record the sequel: Dr George Williamson was appointed chief executive on 17 February 2026 and joined on 20 May 2026, and its chair, Dr Doug Gurr, announced on 1 April 2026 that he would step down on taking a permanent role at the Competition and Markets Authority.

Talent: a large sector, a net outflow of researchers

The government’s own AI Sector Study 2024, published in September 2025, counted 5,862 AI companies, 86,139 full-time-equivalent jobs, £23.9 billion of revenue and £11.8 billion of gross value added, with revenue up 68 per cent in a year. That is the applied sector, and it is one of the largest outside the United States and China.

The evidence on the researchers who would build a frontier model is thinner and points the other way. UKRI’s 2025 global-mobility report records a small net outflow of internationally mobile scientific authors from the UK, and that a third of non-British researchers surveyed had definite or strong plans to work outside the UK within five years. The 2025 immigration white paper proposed extending the qualifying period for settlement from five to ten years while expanding the Global Talent visa; the government has since said it is “doubling the resourcing of our Global Talent Taskforce” and reimbursing visa fees “in a limited and targeted manner” for skills including AI. No peer-reviewed study of an AI-specific brain drain from the UK was located for this page.

Compute: one good machine, two orders of magnitude short

This is where the gap is largest and most measurable.

The government’s Future of Compute Review (March 2023) found that “as of November 2022, the UK had only 1.3% share of the global compute capacity” and had no system in the world’s top 25. Since then the UK has built Isambard-AI at the University of Bristol: 5,448 Nvidia GH200 chips, about 23 AI exaflops, £225 million, roughly 5 megawatts, operating since July 2025 and ranked 11th in the world at launch. It is a good machine, well used (about 1,000 projects in its first year), and it is the whole of the UK’s frontier-class public compute.

For scale: Epoch AI’s April 2025 survey of AI supercomputers found that “the United States accounts for about three-quarters of global AI supercomputer performance”, China about 15 per cent, and that “traditional supercomputing powers like the UK, Germany, and Japan now play marginal roles”. The largest single cluster it describes, xAI’s Colossus, used up to 200,000 chips and about 300 megawatts. The leading systems’ performance, it found, has doubled roughly every nine months. Isambard-AI is about one-fortieth of Colossus by chip count and one-sixtieth by power.

The government’s stated response is the UK Compute Roadmap (July 2025): up to £2 billion to 2030, a twentyfold expansion of the public AI Research Resource from 21 to 420 AI exaflops, and a national supercomputer at Edinburgh (up to £750 million, building works begun June 2026, due in autumn 2027). Even delivered in full, 420 exaflops would be well short of what a single US developer was operating in 2025, and the roadmap itself describes the UK estate as “fragmented”.

Private compute is where the big numbers have been announced, and where delivery lags. In September 2025 Nvidia pledged “120,000” GPUs for the UK, and Nscale up to 58,640, including a 23,040-chip Microsoft supercomputer at Loughton, Essex, and OpenAI’s “Stargate UK” in the North East. As of 23 September 2026: Stargate UK was paused in April 2026, with OpenAI citing “regulation and the cost of energy”; Microsoft’s own May 2026 account gave no delivery date for Loughton; trade press reports Loughton’s grid connection slipping to 2027. UK AI policy: what the government is actually doing tracks each pledge.

Energy and data centres: the binding constraint

National Grid transmission pylons crossing farmland near Earl Stonham, Suffolk
National Grid pylons near Earl Stonham, Suffolk. Frontier training is an electricity problem before it is anything else.Photo: Geoff Pick, CC BY-SA 2.0, via Wikimedia Commons (cropped)

Frontier training is an electricity problem before it is anything else. Three official or industry findings frame the UK’s position.

  • Price. The Department for Energy Security and Net Zero’s Quarterly Energy Prices (September 2025) found that of the 25 IEA countries reporting industrial electricity prices for 2024, “the UK had the highest price including taxes and levies”. The United States did not report in that table, so the comparison is with Europe and Asia rather than with the country that hosts most frontier compute.
  • Grid. In December 2025 the National Energy System Operator began reforming a connection queue of “more than 700 GW of generation and storage projects waiting for grid access”, prioritising projects “aligned to national energy targets and… ready-to-build”. Data centres wait in that queue.
  • Capacity. CBRE puts London’s new data-centre supply at a record 193 megawatts in 2025 and about 180 in 2026, with London over 80 per cent of the national market, power “limited” in the west-London clusters and vacancy at a record-low 5.9 per cent. The Tony Blair Institute estimated in August 2025 that the UK had about 1.8 gigawatts of capacity against a need of 6 gigawatts of AI-ready capacity by 2030.

The government’s answer is the five AI Growth Zones, each required to show it can support at least 500 megawatts, and a preferred bidder (Rolls-Royce SMR, June 2025) for small nuclear reactors whose first site, Wylfa, has no announced first-power date. A freedom-of-information exercise reported in August 2026 found a single relevant planning permission granted in 2025 in the district containing the first zone.

Chips: design without fabrication

The UK’s chip position is unusual: world-class design, no advanced manufacturing, and foreign ownership of the design.

Arm, headquartered in Cambridge, licenses the processor architecture used in the great majority of the world’s smartphones and, increasingly, in data-centre chips. It is listed on Nasdaq and majority-owned by SoftBank of Japan. On 24 March 2026 it announced, in its own words, that it was expanding “to silicon products in historic company first”, and on 8 September 2026 it launched a data-centre processor it calls the “AGI CPU”, a product name, not a claim about artificial general intelligence. Those are changes of business model, not of ownership. Graphcore, the Bristol AI-chip designer, was bought outright by SoftBank in July 2024 and has since received more than $450 million from its owner while, in the reporting of May 2026, planning a £1 billion campus in Bengaluru. The UK has no leading-edge fabrication plant; its most advanced fab, Newport Wafer Fab, was sold by Nexperia to Vishay for $177 million in 2024 after a government divestment order. The UK’s publicly documented AI compute (Isambard-AI on Nvidia chips, Dawn on Intel with an AMD upgrade planned, the announced Nscale and Microsoft clusters on Nvidia) runs on imported accelerators, and Britain has no leading-edge fabrication capacity to make them.

That also makes the UK a dependant of US export policy. The January 2025 US “AI diffusion” rule placed the UK in the least-restricted tier; the US Department of Commerce rescinded that rule on 13 May 2025 and said it would “issue a replacement rule in the future”. As of 23 September 2026 this page had not identified a published replacement rule; the UK’s position under any successor is therefore unstated here.

Capital and ownership: British money, American owners

British AI companies raised $12.6 billion in the first half of 2026, four times the same period a year earlier and 39 per cent of all European AI venture capital, according to Dealroom. That is the strongest capital market in Europe. It is also small against the country that matters: the 2026 AI Index puts US private AI investment in 2025 at over $344 billion.

Ownership follows the money. DeepMind belongs to Alphabet; Arm and Graphcore to SoftBank; Wayve’s largest backers are SoftBank, Nvidia and Microsoft; the London research offices of OpenAI, Anthropic and Microsoft AI report to California and Washington State. The government’s £500 million Sovereign AI fund, deploying since April 2026, is the first public attempt to hold equity in British AI companies; its first investments are early-stage. The £31 billion of US technology pledges announced in September 2025 sat inside a “Tech Prosperity Deal” that the United States paused in December 2025 and that a Commons committee described in July 2026 as “a non-binding framework for co-operation” that “was suspended before it could produce substantive outcomes”.

What the UK’s own institutions say

None of the official documents sets AGI or superintelligence as a national goal, and none claims the UK could build one alone. The Action Plan’s language is about being “an AI maker, not an AI taker” and having “companies at the frontier that will be our UK national champions”. The Compute Roadmap defines sovereignty modestly: “ensuring we have the ability to act independently and effectively where it matters most – to allocate compute to national priorities”. The House of Lords Communications and Digital Committee has twice (February 2024; March 2026) warned of “long-term dependence on opaque models trained overseas, with most benefits accruing to a small number of US-based firms”, and noted that “large-scale data collection, processing and model training, typically occur outside the UK”. The Tony Blair Institute’s August 2025 advice was that the UK should concentrate on “demonstrating to the world how to effectively apply” AI rather than on frontier training. The AI Security Institute’s Frontier AI Trends Report says only that “it is plausible that in the coming years, this trend may lead to capabilities widely acknowledged as Artificial General Intelligence (AGI)”, a statement about the technology, with no country attached.

The balance sheet

Capability What the UK has (evidence) What constrains it (evidence)
Frontier laboratories DeepMind founded, headquartered and led from London, AGI its stated aim; OpenAI’s and Anthropic’s largest overseas research offices; Microsoft AI hub All owned by US companies; Google’s TPU regions include no UK zone; no UK frontier-model training cluster publicly identified by any of them (as of 23 September 2026)
Model output Epoch (June 2024): 42 large-scale models cumulatively AI Index 2025: no UK “notable” models in 2024, DeepMind’s counted as American
Research base 12 AI doctoral centres (£117m); Turing fellowships; top-4 in Tortoise index; 7 of top-100 cited papers (2023) Alan Turing Institute restructured under ministerial instruction 2025–26
Talent 5,862 AI companies, 86,139 jobs (2024); expanded Global Talent visa Net outflow of research authors; a third of non-British researchers considering leaving; settlement period extended
Public compute Isambard-AI: 5,448 chips, ~23 AI exaflops, 5 MW; roadmap to 420 exaflops by 2030 US ~75% of AI supercomputer performance; UK “marginal”; Colossus ~200,000 chips, ~300 MW
Private compute Pledges: Nvidia 120,000 GPUs; Nscale 58,640; Microsoft 23,040 at Loughton Stargate UK paused; Loughton undated; grid slippage reported
Energy Five Growth Zones at ≥500 MW each; SMR bidder chosen Highest industrial electricity price of 25 reporting IEA countries (2024); 700 GW grid queue; ~1.8 GW capacity vs 6 GW estimated need
Chips Arm architecture; Graphcore design Both SoftBank-controlled; no advanced fab; dependence on Nvidia/TSMC and US export policy
Capital $12.6bn H1 2026; 39% of European AI VC; £500m Sovereign AI fund US private AI investment >$344bn (2025); Tech Prosperity Deal paused
Stated goal “AI maker, not taker”; “national champions” No official AGI or superintelligence objective

What the evidence does and does not settle

The evidence settles the narrow question. On the published evidence, Britain could not today run a frontier-scale training run on hardware, power and money located in Britain and owned in Britain: the documented compute is one to two orders of magnitude short, the electricity is the dearest in the reporting group, the accelerators are imported, and the owners are elsewhere. The evidence does not settle the wider one. Whether AGI or superintelligence can be built at all, by anyone, is an open question in the field; whether a country that hosts the research staff of three frontier laboratories but none of their hardware counts as a builder is a matter of definition. What can be said, with dates, is what the UK owns, what it rents, and what it has announced. That is what this page will keep updating.

Sources

  1. Google DeepMind, About, checked 23 September 2026; “Taking a responsible path to AGI”, 2 April 2025.
  2. Google Cloud, “Google opens Waltham Cross data centre…”, 16 September 2025: Hassabis quotation; Cloud TPU regions and zones, checked 23 September 2026.
  3. Stanford HAI, AI Index Report 2025, chapter 1, April 2025; IEEE Spectrum, “State of AI Index 2026”, April 2026: US private investment.
  4. Epoch AI, “Trends in AI supercomputers”, 23 April 2025; “Large-scale models by country”, 19 June 2024.
  5. OpenAI, “Introducing OpenAI London”, 28 June 2023; MLQ, “OpenAI establishes first permanent office in London”, 13 April 2026; eWeek, “Anthropic London expansion”, 17 April 2026; Microsoft, “Announcing new Microsoft AI hub in London”, 7 April 2024.
  6. Wayve, Series C, 7 May 2024; Synthesia, Series E, 26 January 2026; ElevenLabs, Series D, 4 February 2026; Isomorphic Labs, $600m raise, 31 March 2025.
  7. Tortoise Media, The Global Artificial Intelligence Index 2024, 19 September 2024.
  8. UKRI, “UKRI invests in the next generation of AI innovators”, November 2023; “World-leading researchers named as Turing AI fellows”, 23 October 2024; Global mobility evidence report 2025.
  9. Alan Turing Institute, “£100m investment…”, 6 March 2024; news: Dr George Williamson appointed CEO, 17 February 2026, joined 20 May 2026; Dr Doug Gurr to step down as chair, 1 April 2026; Civil Society, “Alan Turing Institute CEO resigns amid government pressure”, 5 September 2025: the 2025 events, as reported.
  10. DSIT, Artificial intelligence sector study 2024, 3 September 2025.
  11. House of Commons Library, Changes to legal migration rules, updated 21 September 2026; written ministerial statement HLWS1262, 20 January 2026.
  12. DSIT, The future of compute: report of the review of the independent panel of experts, March 2023; UK Compute Roadmap, July 2025; AI Research Resource, updated 25 August 2026.
  13. University of Bristol, Isambard-AI launch, 17 July 2025; DCD, “UK’s 5MW Isambard-AI supercomputer goes live”, 17 July 2025; EPCC, turf cut, 25 June 2026.
  14. Nvidia, “NVIDIA and United Kingdom build nation’s AI infrastructure…”, 16 September 2025; Nscale, UK AI infrastructure announcement, 16 September 2025; Microsoft UK, “Building the foundations of the UK’s intelligence economy”, 28 May 2026; Data Centre Magazine, “Stargate UK: why is OpenAI halting…”, 10 April 2026.
  15. DESNZ, Quarterly Energy Prices, September 2025, 25 September 2025; NESO, “NESO implements electricity grid connection reforms”, 8 December 2025; CBRE, UK Real Estate Market Outlook 2026: Data Centres; The Register on the Tony Blair Institute report, 4 August 2025; Data Centre Review, FOI on Growth Zone planning, 20 August 2026; Rolls-Royce SMR, Wylfa confirmed, 13 November 2025.
  16. Arm, IPO pricing, 13 September 2023; Q3 FYE26 results, 4 February 2026; newsroom: “Arm expands compute platform to silicon products in historic company first”, 24 March 2026, and “Arm expands AI infrastructure for the agentic era with AGI CPU and Neoverse CSS N4”, 8 September 2026; Graphcore, “Graphcore joins SoftBank Group”, 11 July 2024; Tech Funding News on Graphcore, 13 May 2026; Nexperia, sale of Newport Wafer Fab to Vishay; US Department of Commerce BIS, rescission of the AI diffusion rule, 13 May 2025.
  17. Dealroom, “UK AI startups raised $12.6B in H1 2026”, 7 August 2026; Business and Trade Committee, report on the Tech Prosperity Deal, 4 July 2026.
  18. DSIT, AI Opportunities Action Plan, 13 January 2025; House of Lords Communications and Digital Committee, LLMs report, 2 February 2024, and AI and copyright report, 6 March 2026; AI Security Institute, Frontier AI Trends Report, December 2025; FCDO, UK statement at the UN Security Council, 24 September 2025.

Common questions

Is the UK trying to build AGI or superintelligence?
Not as a stated national goal. As of 23 September 2026, no UK government document sets AGI or superintelligence as an objective. The AI Opportunities Action Plan (January 2025) speaks of making Britain "an AI maker, not an AI taker" and of "national champions" at the frontier, and the government's compute roadmap defines sovereignty as the ability "to allocate compute to national priorities". Google DeepMind, headquartered in London, does state AGI as its aim, but it is owned by Alphabet, a US company.
Does the UK have the computing power?
Not at frontier scale. Its largest public AI computer, Isambard-AI in Bristol, has 5,448 Nvidia chips and draws about 5 megawatts; the largest private training clusters in the United States in 2025 had 100,000–200,000 chips and drew about 300 megawatts. Epoch AI estimates the United States holds about three-quarters of global AI supercomputer performance and describes the UK's role as "marginal". Google publishes no UK zone for the TPU hardware it trains its models on.
What does the UK have going for it?
The founding home and headquarters of Google DeepMind; the largest overseas research offices of OpenAI and Anthropic; Arm, whose chip designs are in most of the world's phones; universities that rank highly in AI research; an AI sector the government counted at 5,862 companies and 86,000 jobs in 2024; and the largest share of European AI venture capital: $12.6 billion in the first half of 2026, 39 per cent of the European total.