The Digital Sovereignty Brief #6 — "DSIT is abolished"
The department that had just asked whether Britain should hold a position on the sovereignty of its own data was abolished this week, its digital and AI functions split three ways across Whitehall.
The Digital Sovereignty Brief · Issue #6 · Friday 24 July 2026 A weekly read on UK digital sovereignty: who owns the tech, where the data lives, and where it could be cut off.
The week in six lines:
The government abolished DSIT. On 21 July the Prime Minister, Andy Burnham, closed the Department for Science, Innovation and Technology and split its functions across three departments — one week after DSIT had opened a call for evidence asking whether Britain should hold a position on the sovereignty of its own data.
AI policy moves to the centre; the rest scatters. AI strategy and the AI Security Institute go to the Cabinet Office, which gains a new AI taskforce and a Prime Ministerial adviser on AI; the Government Digital Service, telecoms and digital-identity policy go to an enlarged DCMS. Where data policy lands, the government has not said.
A US defence investor is buying a British defence-optics maker. Washington’s Arlington Capital agreed a £345.6m take-private of Gooch & Housego — a photonics firm supplying aerospace and defence, and a producer of germanium components — conditional on a UK national-security clearance.
The statistics regulator qualified the NHS’s effectiveness claims for its national data platform. NHS England has caveated the figures, and agreed to publish trust-level data and commission an independent evaluation. (The platform is the one built by Palantir — a link we make, not one the regulator drew.)
The industry’s own booster report concedes the value leaves. Tech Nation’s 2026 report puts 57p of every pound of AI-attributed UK exit value going to US acquirers, and 9p staying in Britain.
And the ONS put a number on the dependency: UK business AI use has tripled to about 35% since 2023, adopted mostly through other people’s software.
The department that had begun, however tentatively, to name Britain’s technology dependency was dissolved this week, and its digital, data and AI functions dispersed across three centres of government. In the same fortnight a US defence house moved to buy a British defence-optics supplier, the national statistician measured how fast and how shallowly the country is wiring itself to off-the-shelf AI, the sector’s own promoters conceded that most of the value of UK AI leaves for America, and the statistics regulator qualified the case for the US-linked platform holding NHS data. The reorganisation may prove sensible administration. It changes none of the week’s other developments.
Ownership & control
A US defence-focused investor is buying one of Britain’s specialist defence-optics makers, and the national-security test now applies. On 16 July Arlington Capital Partners — a Washington DC private-equity firm built around aerospace, defence and regulated industries — announced a recommended £345.6m cash take-private of Gooch & Housego, the AIM-listed photonics group founded in 1948. G&H makes mission-critical optics for aerospace and defence, and produces germanium components the company itself describes as “critical to a number of aerospace & defence and semiconductor applications”, across eleven sites in the UK, US and Asia. The offer — 1,234.9p a share including an interim dividend, a premium of about 41% to the pre-announcement close — is recommended by the board and conditional on two clearances: a US antitrust review, and clearance under the UK’s National Security and Investment Act. Control would pass to Arlington Capital Partners VII; the ultimate decision-making sits in Washington.
Why it matters: this is the national-security test working as designed rather than a story of something slipping through. A British supplier of defence optics and a strategic material passing to US-controlled ownership is exactly the kind of transaction the National Security and Investment Act exists to screen, and the deal is explicitly gated on that screen. The test plainly applies. What the coming months decide is what the government judges it protects — the jobs and the sites, the customer relationships, or control of the germanium and the optics themselves. Gooch & Housego / Arlington announcement.
Briefly: the pension consortium chasing IP Group — the FTSE 250 investor holding stakes in Oxford Nanopore and First Light Fusion — let its own deadline pass. Rather than a firm bid or a walk-away by 20 July, Railpen’s consortium had a revised proposal rejected by the board and secured a seven-day extension from the Takeover Panel, to 5pm on Monday 27 July. The rejected terms implied about 71p a share before a contingent element; the board considers that a material undervaluation. If the consortium does firm up an offer, a portfolio of British deep-tech stakes would pass to UK pension capital — the rare ownership move that runs in Britain’s favour. IP Group’s statement.
No new final orders were published under the National Security and Investment Act this week; the most recent remains 1 July. Call-ins are not published, so this covers only what the government puts on the record.
Data & infrastructure
Microsoft’s EU Data Boundary has not moved in over three months. The change log has carried no new entry since 13 April. The most recent substantive entries remain outward-facing — Copilot features added in April to the services that transfer some data out of the boundary, and Anthropic’s models added in January to the continuing transfers that apply across it.
Why it matters: the EU Data Boundary is the most-cited residency commitment in European enterprise software, and it was never a UK commitment in the first place. A commitment that is quiet is not necessarily weakening. But the only recent movements have added services to the transfer-out list, not removed them, which is reason enough to keep reading it closely. Change log.
Otherwise a quiet fortnight on this front: no UK or EU residency commitment, sub-processor change or new region announced in the window. (Specialist outlets reported a 16 July AWS CloudFront fault, originating in a single Frankfurt availability zone, that briefly affected UK-facing services.)
Reports & analysis
The ONS has measured how fast, and how shallowly, Britain is wiring itself to AI. In a release on 20 July, the Office for National Statistics reported that self-reported AI use among UK businesses with ten or more employees has risen from about 12% in late 2023 to about 35% in June 2026 — nearly a tripling — while the depth of that use has barely moved: the average adopter uses 1.6 AI technologies, up from 1.4, and only one adopter in ten reports using AI “extensively”. Large language models are the most common technology, used by 18% of businesses. The detail that matters for sovereignty sits in how firms adopt: the two most common routes, across every industry, are free-to-use software and purchased off-the-shelf tools, with in-house development far rarer. The ONS notes, drily, that free-to-use software has “no market price”, which makes the dependency partly invisible in the national accounts.
Why it matters: the ONS does not name the vendors, and we should be precise that the next step is our reading, not the statistician’s. But the most-used categories — general-purpose language models and visual-content tools — are dominated by a small number of US-controlled providers, and the dominant adoption routes are precisely the ones that involve no British software and no negotiated contract. What the figures describe, in neutral language, is a fast-growing dependence formed by default: adopted quickly, held shallowly, and wired to tools the adopter neither owns nor can readily leave. ONS: Artificial intelligence in UK businesses.
Parliament’s own librarians have mapped the dependency. Ahead of a Lords debate on 23 July, the House of Lords Library published a briefing, “Digital and technology policy and national sovereignty”, that assembles the picture in one place: the three US hyperscalers holding close to two-thirds of the global cloud market; the reach of US law over data held by US-controlled providers, including a Microsoft executive telling the French Senate last year that he could not guarantee French data would never be passed to US authorities; the roughly £1.6bn the UK has committed to sovereign AI; and named voices on both sides of the argument, from OpenUK’s Amanda Brock — who holds that openness, not ownership, is the right test — to those who call data and AI sovereignty existential.
Why it matters: an impartial parliamentary briefing is a useful marker of where the mainstream of the argument now sits, and this one treats digital sovereignty as a settled subject for scrutiny rather than a fringe concern. Two cautions on the numbers it collects: the cloud-market share is a research-firm estimate, and the widely cited figure that only 35% of UK organisations have full clarity on where their data sits comes from a UK cloud vendor with an interest in the answer. The briefing’s value is the synthesis and the range of named opinion, not any single statistic. Lords Library briefing.
Even the industry’s own promoters concede the value leaves. Tech Nation’s 2026 report, “The Next Wave of UK AI” — a document carrying forewords from the then-Chancellor and the then-DSIT secretary of state, produced by Founders Forum Group to sell Britain as a place to build and fund AI companies — contains a set of figures that cut against its own purpose. On the report’s data (from Dealroom and its mapping partner Thema), US-based investors account for 53% of AI investment into UK companies; US buyers made 46% of UK AI acquisitions in 2025; and, on Thema’s bespoke measure that attributes exit value by AI workforce, 57p of every pound of value from UK AI exits goes to US acquirers, against 9p that stays in the UK.
Why it matters: the market-sizing headlines in a promotional report are worth little, and we are not repeating them. But a concession against interest is worth more than a claim that serves it. The value-capture figure rests on Thema’s proprietary method, which we would want to see tested; taken at its own word, though, it describes Britain as a good place to start an AI company and a poorer place to keep the proceeds of building one — a finding that carries more weight for appearing in the industry’s own prospectus. Source: Tech Nation, The Next Wave of UK AI (2026).
Regulation & policy
The government abolished the department that had just begun to ask the sovereignty question. On 21 July, in a machinery-of-government change, the Prime Minister closed the Department for Science, Innovation and Technology and redistributed its work. The government’s own confirmation is on the DSIT page itself, which now reads: “This organisation is changing. It’s being replaced by the Department for Business, Innovation, Science and Trade, the Department for Digital, Culture, Media and Sport and the Cabinet Office.” The Cabinet Office’s machinery-of-government fact sheet sets out the moves that are settled: AI strategy, public-sector AI adoption and the AI Security Institute go to the Cabinet Office, where a new AI taskforce and a Prime Ministerial adviser on AI now sit; the Government Digital Service, telecoms, media, online safety and digital-identity policy go to an enlarged DCMS; and the Investment Security Unit moves into the new business department under Jonathan Reynolds.
Why it matters: concentrating AI policy at the centre of government is defensible, and may even strengthen it. But the same document is silent on where several functions this publication follows most closely now sit — most pointedly, data policy, including the “data flows” call for evidence DSIT opened on 15 July, which asked industry whether the government should take a position on the sovereignty of UK data and remains open until September. The department opened that question on 15 July. Six days later it was abolished, and the fact sheet does not name who now owns the question. The machinery for a sovereignty position is mid-reorganisation, and where it lands is not yet on the record. GOV.UK: DSIT. · Machinery of government fact sheet.
The statistics regulator qualified the NHS’s effectiveness case for its national data platform. On 22 July the Office for Statistics Regulation closed a review of how NHS England presents the benefits of its Federated Data Platform — the national system for linking NHS datasets. The claims at issue are NHS England’s own headline figures, repeated by ministers: that hospitals using the platform delivered 110,000 additional operations and a 15% reduction in discharge delays. NHS England has now agreed to caveat the figures that rest on before-and-after comparisons, acknowledging that it “cannot draw conclusions about cause and effect as other variables have not been controlled for”; to commission an independent academic evaluation from Imperial College; to publish disaggregated data trust by trust; and to make clear that the case studies on its site are written by the organisations using the platform and are “not authored or verified by NHS England”. The review followed reporting by the Financial Times, which surfaced the quietly-added caveat and found errors in the discharge-delay data; the regulator said it would monitor figures cited by ministers about the platform’s effectiveness.
Why it matters: the platform is the one built by Palantir, the US company whose place in the NHS this publication has followed, and the new government inherits an early decision on the £330m contract. The regulator’s finding does not touch the vendor question directly — it is about how a public body presented its own statistics — but it removes a prop from under the political case. An effectiveness claim that has been formally qualified is a weaker foundation for renewing a contract, and a stronger one for the argument that the decision should rest on evidence that has been independently checked. Office for Statistics Regulation. · FT (paywall).
The Cyber Security and Resilience Bill, which had its Lords second reading on 14 July, now has its committee stage listed for 1 September, after the summer recess. We will pick up the digital-sovereignty amendment argument then. Bill stages.
Briefly, from Brussels: the EU’s digital chief, Henna Virkkunen, told the Financial Times on 21 July that AI has become “a geopolitical weapon”, and that Europe must build its own alternatives to US models faster “or risk being cut off from strategic capabilities”. She cited June’s US export controls on Anthropic’s models — the episode this publication has tracked since June — as evidence of the risk, and said the EU is pursuing both homegrown providers (the model-maker Mistral, the cloud firms Scaleway and OVHcloud) and a “trusted partner” arrangement with Washington to preserve access. It is the same cut-off risk we map for the UK, named from Brussels by the official in charge of Europe’s response to it. FT (paywall).
For detailed analysis of digital sovereignty issues and how they affect the UK visit the Information Matters website.