In January 2025, Nscale announced it had "confirmed" its first UK data centre: a site in Loughton, Essex, purchased and scheduled to go live by the end of 2026, housing up to 45,000 Nvidia GPUs. The announcement was framed explicitly as aligning with the government's AI Opportunities Action Plan, and ministers repeated the framing for over a year afterwards. Fourteen months later, a Guardian reporter drove out to look at the site. It was a scaffolding yard. Land Registry records showed the site was still held by a different company entirely.
That gap, between the announcement and the field, is the whole story. Everything else is detail.
Nscale is not obscure. Spun out of Arkon Energy, an Australian crypto-mining outfit that Josh Payne and Nathan Townsend wound down as bitcoin economics soured, Nscale rebuilt itself around AI data centres and became, within two years, the government's preferred proof that Britain could compete in the compute race. It has genuinely raised money: a Series B, a $2bn Series C at a $14.6bn valuation, backers including Nvidia, Dell, Aker, Citadel and Point72. None of that is in dispute.
What is in dispute is what the government did with the claims. Nscale's own release did not say it intended to invest; it said the investment was confirmed, the site purchased, the jobs coming. That language sat in ministerial statements and DSIT press material for over a year before anyone checked it against the ground.
Nine months after the Loughton announcement, the government needed something to put alongside Donald Trump during his September 2025 state visit. It produced Stargate UK, and this time the press release was the government's own: OpenAI, Nvidia and Nscale, up to £30bn of investment, 8,000 GPUs at Cobalt Park in the North East by the first quarter of 2026, scaling to 31,000.
A Guardian FOI request later established that neither OpenAI nor Nscale had held a single recorded meeting with the relevant local authorities before the announcement. Nvidia met the North East Combined Authority; the other two did not. Of the headline £30bn, roughly £20bn turned out to be a notional figure representing what it would cost to build the full 1.1GW of capacity eventually, not money any investor had committed. Spotlight on Corruption's assessment was blunt: the government's own explanation showed the figure reflected projected cost, not pledged investment.
North Tyneside's Conservative group leader learned about the project from the announcement itself, telling the Guardian he was not made aware of any of the discussions beforehand; he described the arrival of considerable fanfare with no prior consultation and, in his view, no local infrastructure to support it.
OpenAI paused Stargate UK in April 2026, citing UK industrial electricity costs running at roughly four times US and Nordic rates, plus unresolved regulatory questions. A second Guardian investigation in July established that OpenAI had never conducted a site visit at Cobalt Park before the launch event. The flagship announcement of Britain's AI strategy, made to coincide with a head of state's visit, had been assembled without anyone from the lead company setting foot on the site.
The load-bearing fact in all of this is not the scaffolding yard. It is what the Department for Science, Innovation and Technology, as it then was, said when the Guardian asked it to explain the £1.9bn Nscale contract the government had described in its own press release. There was no contract. DSIT's own characterisation was that the commitment amounted to an intention to commit capital; the department conceded it was not actively auditing any of these figures. The numbers behind the UK's AI Opportunities Action Plan, and behind ministers' repeated citation of over £100bn in private AI investment since taking office, come from company self-reporting: unverified, unaudited, taken at face value.
Professor Cecilia Rikap of UCL's Institute for Innovation and Public Purpose had a name for it: phantom investments. Her point was not that Nscale is fictitious. It is that a government hungry for good news has strong incentive to accept generous framing from companies who benefit from being seen as generous.
Nscale had reason to be generous. Guardian analysis of share allocations from October 2025, priced at one pence, found they were worth something in the region of a 350,000 per cent return once the March 2026 funding round landed. A company built on a crypto-mining wind-down, whose predecessor entity's own 2024 accounts flagged significant doubt about its ability to survive, had become, on paper, one of the most valuable start-ups in Europe; propped up substantially by government-adjacent credibility it had not yet earned on the ground.
Loughton finally got a planning application in early 2026, filed only after Guardian inquiries began. The live date has since slipped to the second quarter of 2027. Construction is reported as under way, though not independently verified at time of writing.
Nscale is now marketing a US listing, targeted as early as September 2026, at a valuation approaching $25bn, built on a claimed $51bn contract backlog against roughly $100m of actual second-quarter revenue. No S-1, F-1, or any registration statement has been filed with the SEC. The same pattern that characterised the UK government claims, headline figures well ahead of anything audited or contractually locked, is being carried directly into the IPO pitch; this time for American institutional money rather than British political capital.
Parliament's response so far has mostly been to keep citing the numbers. The day after the Guardian's March story broke, a minister told the Commons that Nscale's role in the AI growth zones was, in part, a result of the government's own support and convening; one way of describing a company whose central UK commitment had just been shown to be uncontracted and unaudited.
This is the Theatre Pulldown in miniature: an institutional claim of operational reality, built for a political moment, that collapses the instant anyone checks it against a land registry, a planning portal, or a site visit. The claim does not need to survive contact with reality. It only needs to survive the news cycle it was built for. By the time the scaffolding yard gets photographed, the announcement has already done its job.
Three weeks before publication, DSIT stopped existing. New Prime Minister Andy Burnham dissolved it in his first Cabinet reshuffle, on 20 July 2026, splitting its functions between an enlarged Department for Business, Innovation, Science and Trade and a reconstituted Department for Digital, Culture, Media and Sport. The department that wrote the January 2025 material, staged the September 2025 Trump-visit announcement, and told the Guardian in March 2026 that it held no contract and audited none of its own figures, no longer exists in any form that could be asked to answer for those things.
In its place sits a single AI minister, Kanishka Narayan, previously a junior minister inside DSIT, now with a seat at the Cabinet table but jointly housed across the Cabinet Office and DBIST rather than in a department of his own. The institutional structure that made, and then failed to substantiate, Britain's headline AI investment claims has been replaced by a person with a title and no department. Whether that produces clearer accountability, or simply gives everyone a new reason to say it is somebody else's inherited mess, is at time of writing unresolved. A machinery of government change is, among other things, a highly effective way to make yesterday's questions somebody else's problem.
Cross-reference: The Theatre Pulldown · The Control Plane Trap
The Sovereign Auditor covers digital sovereignty, cybersecurity governance, and data protection policy—with particular focus on Isle of Man jurisdiction and Crown Dependency issues.
Payments via PayPal. Credentials delivered by email. No Substack. No Stripe. No middlemen.
