Microsoft walked away from Loughton — but the grid may have got there first

Microsoft has withdrawn from a planned UK data centre agreement with Nscale concerning the proposed £2bn facility in Loughton, Essex, where it was expected to become anchor tenant. The Telegraph, which first reported the withdrawal on Sunday, did not establish a specific reason. What it did report is that the project has faced delays and difficulties securing electricity supplies, and that authorities told Nscale the facility might not receive a grid connection until the 2030s. That single reported detail deserves more attention than the corporate choreography around it. If a site of this scale, backed by this kind of tenant interest, cannot be confident of power before the following decade, then the constraint on British AI is not demand, capital or political enthusiasm. It is the grid.

The Loughton project was announced last year with a £2bn pledge and was due to receive its power supply at the end of the year. That timeline has evidently slipped, and the reported connection date now sits in the 2030s. Nscale says it remains committed to both the Loughton project and the UK, is exploring alternative power sources for the site, and points to other expansion sites in the UK. It also says its relationship with Microsoft remains strong and long-term. Microsoft declined to comment. All of that is reasonable corporate positioning, and none of it changes the arithmetic: a data centre without a connection is a building with expensive intentions.

The concentration problem compounds the power problem. According to its IPO filing, Microsoft and Anthropic account for roughly 85% of Nscale's contracts, and the company is seeking to raise as much as $3bn in an IPO as soon as October. A customer base that narrow makes any single withdrawal material, whatever the cause. Separately, Nscale's flagship site in Norway is about six months behind schedule, with only 6.4MW of its planned 30MW capacity operational. That is not a UK-specific story, but it is a reminder that execution risk in this sector is not confined to one jurisdiction or one regulator.

The strongest case against my argument is straightforward, and it should be taken seriously. Microsoft's withdrawal could reflect commercial or strategic factors entirely unrelated to power. The Telegraph did not establish a reason, and no outlet has. It is possible that Microsoft reassessed its UK footprint, reprioritised other markets, or changed its view of the site for reasons that have nothing to do with a connection date. Blaming the grid on the basis of a single unconfirmed cause risks overstating the case, and it would be wrong to assert a motive that the reporting does not support. A sceptic could reasonably say that the grid angle is a convenient narrative imposed on an ambiguous event.

That scepticism is fair, but it does not rescue the underlying point. Even if Microsoft's decision had nothing to do with electricity, the reported 2030s connection timeline stands on its own. It is not a claim about Microsoft's motives; it is a claim about the physical and regulatory capacity of the system that any UK AI tenant must rely on. The question is not whether the grid caused this particular exit. It is whether the grid can support the pipeline that ministers and investors keep describing. On the evidence reported, the answer is not obviously yes.

This matters beyond one company. Nscale's IPO ambitions, its valuation target of up to $3bn, and its ability to find a replacement anchor tenant for Loughton all now sit alongside an unresolved power question. The open questions are real: why exactly Microsoft terminated its involvement; whether Nscale can secure alternative power sources and on what timeline; how the exit affects the IPO; whether a replacement tenant emerges; and what a 2030s connection timeline means for other UK AI data centre projects. None of those questions is answered by treating the episode as a single deal gone wrong.

Policy should follow the constraint. If connection timelines run to the 2030s, then announcements about AI investment, sovereign compute and attracting US technology firms will be paced by substations, transmission capacity and planning, not by press releases. That means treating grid capacity as the binding constraint in industrial strategy, and being honest that capital can arrive far faster than electrons. The alternative is a queue of well-funded projects waiting for a plug that is not coming this decade. Microsoft's withdrawal may or may not have been about power. Britain's AI future, on the reported evidence, almost certainly is.