Scotland: the 9.5 billion euro AI data centre promised "100% renewable" never will be

What the Guardian actually revealed

The site is at Airdrie, in Lanarkshire, east of Glasgow. It is one of the five “AI Growth Zones” designated by the British government to accelerate AI data centres on home soil, alongside Culham (Oxfordshire), the north-east of England and two Welsh zones. Designated on 29 January 2026, the Lanarkshire project is the largest of the five: 8.2 billion pounds of announced investment and 3,400 jobs promised.

On paper, DataVita says it can power the site with more than 1 gigawatt of renewable energy: 400 MW of solar and 800 MW of wind, presented as “energy parks” connected directly to the data centres — the equivalent, according to the figures put forward, of the consumption of around 800,000 Scottish households.

The problem, documented by the Guardian from freedom of information requests: that energy does not exist, and nothing indicates it will exist in the near future. DataVita’s two data centres already in operation, at Glasgow and Chapelhall, currently draw around 24 MW from the conventional grid — that is less than 3% of the gigawatt promised. An analysis drawing on figures from the Scottish association Action to Protect Rural Scotland (APRS) estimates that between 40 and 100 km² of land would be needed to produce such a volume of renewable electricity. Yet DataVita holds only around 4 km² of land (a little over 1,000 acres), and the only concrete project filed to date — an “energy park” led by the property group HFD, which owns 100% of DataVita, providing for up to 19 wind turbines — would cover only around 5% of the announced production.

The government and the developers themselves knew. According to the documents obtained by the Guardian, they were already acknowledging internally that the site had a “power provision issue”. In February 2026, Scottish First Minister John Swinney wrote to DataVita that “power provision remains a key issue, and we will continue to engage on this” — a muted phrase that, six months later, takes on its full meaning. An energy sector consultant quoted by the Guardian sums up the difficulty bluntly: “Going from nothing to the country’s largest onshore wind farm in four years is ambitious.” Asked about it, DataVita makes delivery of its renewable programme conditional — “final commercial agreements, permits, grid connection and consenting processes” — a reply that disputes none of the published figures.

In other words: to date, not even a grid connection has been confirmed for this site, while the AI Growth Zones’ own eligibility criteria require a credible trajectory towards energy self-sufficiency before any designation.

Newarthill, the village that believed in the cheque

In Newarthill, a small village right next to the site, January’s announcement was first welcomed as a windfall. Representatives of a company linked to the project, Oakes Energy Services, canvassed residents door to door to present a future solar farm tied to the complex: free solar panels, tree planting, or outright purchase of properties.

Six months later, hope has given way to anxiety. Residents now fear having to sell their homes or seeing green belt land disappear, with no certainty that the promised jobs and investment will ever materialise. The project also provided for a community fund of 543 million pounds (around 630 million euros) for the area. The Guardian found that, to date, there is no actual money deposited in that fund: its financing is supposed to come from DataVita’s future revenues — provided it generates any.

A village agreeing to lose farmland and green belt in exchange for the promise of a cheque that does not yet exist: that is, at local scale, exactly the same mechanism as the one observed across the whole project.

A repeating pattern: British AI’s “phantom investments”

This is not an isolated case. In March 2026, an earlier Guardian investigation had already exposed a similar phenomenon elsewhere in the United Kingdom, under the name “phantom investments”. The emblematic example: the Nscale site in Essex, presented as the country’s future largest sovereign AI data centre, for an announced commitment of 1.9 billion pounds (around 2.2 billion euros). In February 2026, that land was still being used… as a scaffolding storage yard. The British government itself confirmed that this was not a formal contract, but a simple “intention to commit capital” — and the planning application was only filed after the Guardian’s investigation began.

CoreWeave itself is no newcomer to this. A previous “investment” of one billion pounds announced by the company in 2024 turned out, in practice, to consist of renting space in data centres that already existed — built in 2002 and 2015 — to install Nvidia chips in them, rather than building anything new. The company declined to disclose the number of jobs actually created.

Asked about it, the British department responsible (the Department for Science, Innovation and Technology) confirmed that it plays “no active role” in auditing these commitments: the figures put forward by companies rest entirely on self-declaration. For researcher Cecilia Rikap, of University College London, the finding is unambiguous: these announcements are “phantom investments”, and technology companies “artificially inflate the job creation and economic impact of their data centres to please governments”. Globally, more than 500 billion pounds of AI investment was announced in 2025 alone — with no independent verification mechanism systematically attached to those figures. Le Recul had already documented a cousin mechanism, applied to space rather than electricity, in our breakdown of SpaceX’s flotation and its space data centre that does not exist yet: a spectacular promise, sold as a done deal, while the product itself remains a mock-up.

The whole of Scotland under electrical strain

The Lanarkshire case is only the best-documented part of a wider problem. According to data compiled by opponents of the project, 24 AI data centre projects are currently filed in the Scottish planning system, representing a combined demand of around 6,200 MW — roughly 1.5 times the peak electricity demand of the whole of Scotland.

The mobilisation is not new. As early as March 2026, the Stop Climate Chaos Scotland coalition sent an open letter to John Swinney calling for a moratorium, while Kat Jones, of Action to Protect Rural Scotland, warned that “large-scale AI data centres are the gravest imaginable threat to Scotland’s climate ambitions, to its rural landscapes and to electricity bills”. Since then, the councils of Edinburgh and East Ayrshire have formally asked the Scottish government to freeze new projects until clear “green data centre” criteria and environmental impact assessments are established. The governing party, the SNP, saw a motion in favour of a moratorium tabled at its own national council. John Swinney has publicly acknowledged that the government is “actively weighing” that option — without deciding.

A useful point of comparison: Dublin has applied a de facto moratorium on new data centres since 2022, and several American states, including Pennsylvania and Illinois, are currently debating similar measures. Scotland is not there yet, but the debate it is going through is anything but isolated: it is the same question — who pays AI’s electricity bill, and who actually benefits — that is already being asked from Dublin to Springfield.

CoreWeave, the American partner building on a mountain of debt

One detail changes the reading of the whole affair: CoreWeave, the American partner meant to secure part of the Scottish site’s financing and technology, is not itself on solid financial ground. The company, specialising in AI cloud and closely tied to Nvidia, posts spectacular growth — 2025 revenue expected between 4.9 and 5.1 billion dollars, up around 300% year on year — but remains loss-making, weighed down by around 1.2 billion dollars of annual interest charges. Its total debt exceeds 10 billion dollars, its total liabilities are close to 29 billion dollars, and its order book reaches 66.8 billion dollars — future commitments, not cash available today. To fund its 2026 capital expenditure alone, up to 30 billion dollars, the company relies on debt rather than equity.

In January 2026, Nvidia had to inject a further 2 billion dollars into CoreWeave, at 87.20 dollars a share — an operation widely read by analysts as a safety net meant to stop its most important cloud partner from wobbling. In other words: the company meant to help deliver a “self-sufficient” data centre in Scotland itself depends, in order to stay standing, on the goodwill of a single chip supplier. Le Recul had already documented this mechanism of decisions taken by a handful of American players, with consequences that then land elsewhere, in our article on OpenAI’s AI dividend that stops at the American border: here, it is no longer “the rest of the world” that inherits the risk, but a Scottish village and its local public finances.

And if the scenario played out elsewhere, including in France?

Nothing in this mechanism is specific to the United Kingdom. Wherever governments seek to attract “gigafactories” or AI growth zones with announced billions, the temptation exists to take at face value the renewable and jobs promises put forward by companies — all the more so since, as the British case shows, the state itself sometimes admits it lacks the means or the will to verify those figures.

France is not immune to the subject: Le Recul had already put figures, in its investigation into the real environmental cost of AI, on the 830 million dollar debt financing announced in March 2026 for a Mistral AI data centre near Paris, at Bruyeres-le-Chatel — a project distinct from the Scottish file, but built on the same ambition of digital sovereignty, and therefore subject to the same questions: what real grid connection, what land footprint, what verifiable jobs. The 3,400 jobs promised in Scotland join an already long list of AI-related job announcements that run, once the figures are audited, into a more modest reality — a pattern Le Recul also follows on the AI-related layoffs side.

So the lesson of Lanarkshire is not only Scottish: it is a reminder that the “renewable” or “sovereign” label stuck on an AI data centre is worth, for now, only as much as the independent verification that comes with it — and that in the United Kingdom as elsewhere, that verification remains, most of the time, still to be invented.

What to take away

On 29 January 2026, the British government designated an AI data centre complex worth 8.2 billion pounds (around 9.5 billion euros) at Airdrie, Scotland, led by CoreWeave and DataVita, with the promise of 100% renewable operation and 3,400 jobs by 2030.

A Guardian investigation, published on 6 July 2026, reveals that this renewable promise is unachievable within the announced timescale: around 24 MW delivered today against 1 GW promised, and a land requirement (40 to 100 km²) far above the land actually held (around 4 km²).

A community fund of 543 million pounds promised to the residents of Newarthill contains, to date, no actual money — its financing depends on DataVita’s future revenues.

This pattern is not isolated: an earlier Guardian investigation, in March 2026, documented comparable “phantom investments” elsewhere in the United Kingdom, notably at Nscale, with no authority systematically auditing these announcements.

Across Scotland, 24 data centre projects already represent around 1.5 times the country’s peak electricity demand, which has pushed several councils to call for a moratorium that is still under consideration.

The figure to remember

3%.

That is, to date, the share of the gigawatt of renewable electricity promised for the Lanarkshire AI data centre that actually exists — around 24 MW delivered out of the 1,000 MW announced. The rest rests, for now, on a planning application not filed, a grid connection not confirmed, and an empty community fund.