What does the Cambridge map say Britain actually holds?
A University of Cambridge audit published on 18 November counts 41 distinct positions linked to UK firms and institutions, spanning 13 of its 14 layers. Of the seven elements that the three UK policy documents it reviewed name precisely, only three match a position Britain holds. Our news report covers the headline findings. The more useful question for UK readers is what kind of leverage those 41 positions give, because the same week the AI minister, Kanishka Narayan, set out an approach that depends on it. Asked whether Britain needs its own frontier model, he told City AM: “I don’t see it as a matter of necessity as long as we have trusted interdependence and access.”
Strategic Reality: Narayan’s case rests on Britain holding positions that other countries depend on. The Cambridge audit maps UK positions element by element, and its answer is narrower than a raw count of 41 suggests.
Why does the gap between holding and naming matter?
The research comes from AIxGEO, a project at the university’s Bennett School of Public Policy. Its policy paper and country report split AI production into 137 elements grouped in 14 layers, with 351 finer sub-elements, and then check 117 official documents against that list. The documents come from 43 national governments plus the EU and the OECD. One element in three is never referred to.
The university’s press release says the UK’s AI policies name “just three” of its 41 elements. The country report is more exact. Its UK corpus is three documents: the Compute Roadmap, the National Security and Investment Act 2021, and the government’s response on AI regulation. Those documents name seven elements precisely, and three of them match held positions: nuclear and small modular reactor capacity, standards, and the export-control and licensing regime. Processor IP, the position Arm gives Britain, is among the 38 held positions left outside precise naming.
| Figure | Value | Strategic implication |
|---|---|---|
| UK positions in the audit | 41 distinct, across 13 layers | Breadth is real, but 38 of the elements come through where suppliers are headquartered, against 14 through where they operate |
| UK policy documents reviewed | 3 | A small corpus; the authors note that how many documents they hold, and how deep they go, differs tenfold between countries |
| Elements precisely named | 7 | Three match UK holdings; none is processor IP |
| Arm’s share of the semiconductor-IP market | Roughly 41% | The report calls it the UK’s clearest difficult-to-replace capability |
| Arm’s ownership | More than 80% SoftBank | Holding a position and controlling it are different things |
Critical Context: The authors warn against reading too much into the naming count. The country report says the UK documents also cover grid access, data centres and evaluation by function rather than by name, so “limited precise naming should not be read as complete policy neglect”. The policy paper goes further: set against Japan, Germany and the Netherlands, it judges that UK policy covers some of what Britain holds “with greater precision”.
Is Britain’s leverage where the minister says it is?
On the evidence of this audit, Britain at least holds positions in each of the areas he names. City AM reports that Narayan picked out three fields for investment that allies would come to rely on: designing chips, compound semiconductors and scientific research. The policy paper’s account of the UK portfolio covers each one. Arm and Imagination Technologies supply processor IP, IQE supplies compound-semiconductor materials, and AISI, Epoch AI and Apollo Research hold positions in evaluation. The paper also credits Linde for industrial gases, Oxford Instruments for chipmaking equipment, Global Marine for submarine cables, and three bodies (UK Biobank, Genomics England and the NHS) for health and genomic data.
The paper’s own verdict is that the UK “tests whether supply-chain power survives where a country manufactures relatively little - and it does”. In its reading, leverage can rest on IP, infrastructure, regulation or who controls access to data, not only on factories.
The catch is in the minister’s own words. “As long as I have insurance on the access, I’m much more focused on piling in bits we can try to build others’ dependence on us,” he said. The Cambridge toolkit gives a sharper way to judge which bits qualify. Its second audit question counts the alternative suppliers for each position, because “a position held by one of two firms worldwide is a bargaining asset; a position held by one of twenty is a market share.” The country report singles out Arm’s processor IP as the UK’s clearest difficult-to-replace capability. That is the kind of bargaining asset the minister’s strategy needs.
Strategic Insight: One of the audit’s questions is whether a single state can use a position on its own. The answer is yes only when the component is made there, the firm that owns it is based there, and that state’s law governs it. Arm shows why the distinction matters.
Does ownership undercut the leverage?
The country report is careful here. Arm is headquartered in Britain, majority-owned by Japan’s SoftBank and listed in the US. Its designs feed chip programmes and foundries abroad, and those depend in turn on equipment and inputs from other countries. The report’s conclusion is that the UK’s contribution is “a qualified intellectual and industrial input within a multinational chain, rather than unilateral authority over every chip using it”. The paper applies the same test to the Netherlands, where the EUV scanner depends on German optics and a light source made by a US-incorporated ASML subsidiary. No country in its cases controls its portfolio completely.
Who is affected by the gap?
Ministers and officials writing the next strategy documents. The audit’s measure is what policy text names, not what it reaches. A pledge to secure the semiconductor supply chain, in the paper’s terms, names a layer rather than a position. Its point is that governments “cannot choose which positions matter” when their own rules leave those positions unnamed. Processor IP and industrial gases are UK positions that the reviewed documents do not name precisely.
UK firms that hold the positions. IQE, Oxford Instruments, Linde and Arm appear in the audit as sources of national leverage. The paper ties each type of position to instruments that can reach it, among them licensing, investment, procurement and standards. The 2021 Act that governs investment screening is one of the three documents in the UK corpus.
Public bodies holding data. The audit treats the health and genomic data held by NHS bodies and research institutions as a UK position in its own right. The government’s top-level independent advisory body on science and technology, the Council for Science and Technology (CST), in a July letter to the minister published on 19 October, called data “one of the UK’s most important assets” and asked government to trial access arrangements in two or more high-value areas, health and creative data among them.
What this means for the sovereignty debate
City AM reports Narayan’s comments against mounting pressure for Britain to depend less on US tech firms. The Cambridge work adds a different lens. The CST letter asked government to map the AI ecosystem first, “including computing power, data, skills, models, governance and public adoption”, and to use the map to decide what Britain should own, share with partners or secure access to. That list is close to the five inputs the Cambridge paper says policy debates typically focus on: “compute, data, models, talent and capital”. The same letter asked government to consider the Alan Turing Institute’s idea of a sovereign base model built in the public interest. The published letter does not mention chips, semiconductors, wafers or packaging at all, the physical middle where the paper finds policy largely silent. Narayan, by contrast, has at least named chip design and compound semiconductors.
CETaS, the Turing’s Centre for Emerging Technology and Security, makes its own case for a British model in a new sovereignty roadmap. It is not a frontier model. The roadmap calls an attempt to match the mostly US and Chinese frontier labs neither realistic nor desirable, and proposes instead a national mission to build a capable open-weight base model of medium size, to serve as a fallback in a crisis and a base for high-stakes work such as defence, security and the national infrastructure the country relies on. Where CETaS and Cambridge meet is Arm. The roadmap names Arm and DeepMind, both started in Britain and now in foreign hands, as evidence that the UK struggles to hold on to AI firms as they grow. Its answer is a Strategic AI Capability designation for a few critical firms, offering faster access to capital, compute and data in exchange for keeping key capabilities in the UK. Cambridge’s audit shows why the question matters: the position it calls Britain’s clearest difficult-to-replace capability belongs to a company that is majority foreign-owned.
Reality Check: CST’s map would be the place to fix this. If it follows the familiar list, it will reproduce the gap Cambridge found. If it works at element level, it could name processor IP and industrial gases as UK positions.
Open challenges
Access is the assumption doing the work. Narayan’s position depends on “insurance on the access”, and he cites AISI’s pre-release access to advanced models as evidence that influence does not need a home-grown frontier system. The Cambridge audit does not test that access. It shows that the UK’s offsetting leverage is real, and that its clearest example, Arm, is majority foreign-owned.
The corpus is small. Three documents is a thin basis for judging UK policy, and the authors acknowledge that their corpus is not exhaustive. Their method counts only what a document states, not what its powers could reach in practice, and they accept that policy framed in operational terms is harder to pick up. The finding is best read as a test of how specific UK strategy is, not of whether government is active.
Leverage is not the same as resilience. The paper says plainly that complete independence cannot be achieved and is not needed. Knowing where Britain holds a scarce position does not tell a public body what to do if a model provider withdraws service. CST frames sovereignty partly as the ability to “maintain resilience under duress”, which is a separate question from bargaining power. CETaS answers it through procurement: critical sectors most exposed to disruption would hold two independent model options, preferably from suppliers in different jurisdictions, and rehearse moving between them.
What to watch
- The CST-recommended AI Sovereignty Plan. If the government produces the ecosystem map CST recommends, and it names positions such as processor IP, industrial gases and compound-semiconductor materials, the Cambridge gap is closing. A map built on compute, data and models alone would confirm it.
- Whether AISI’s pre-release access holds. It is the access the minister relies on. Any narrowing of it would weaken his case and strengthen the argument for the domestic fallback model CETaS wants.
- UK use of the Cambridge toolkit. The fourth of its five questions sorts positions into three lists: those a government both names and holds, those it holds without naming, and those it names without holding. A government audit along those lines, or a departmental response to the report, would show whether Whitehall accepts the measure.
Britain’s leverage in AI is narrower and more specific than either a raw count of 41 or a sovereign model would suggest. The audit supports the minister’s instinct to build on what others depend on. It also shows that the UK documents it reviewed have yet to name most of those positions precisely.
Source: University of Cambridge, A third of the global AI supply chain is ‘invisible’ to governments, 18 November 2026, and the AIxGEO publications behind it: Liu, Turobov, Coyle, Harding and Krook, The invisible AI supply chain (policy paper), and Liu and Turobov, AI supply chain country report, Bennett School of Public Policy. Also drawn on: City AM, UK does not need its own frontier AI model, says AI minister, 18 November 2026, the CST’s letter to the minister, published on GOV.UK on 19 October 2026, and the Alan Turing Institute’s announcement of the CETaS roadmap Delivering the UK’s AI Future: A Sovereignty Roadmap for 2030. See also our news coverage of Narayan’s frontier model remarks, the Turing’s AI sovereignty roadmap and the San Francisco Declaration on AI supply chains, and our analysis of Britain’s sovereign AI funding gap.
This strategic analysis was written by Resultsense, a UK-focused AI news and analysis publication. We will be watching whether the government’s sovereignty map names the positions Britain already holds. Read more analysis at Insights, or get in touch.