Microsoft has withdrawn from its deal for Nscale’s planned Loughton data center in Essex, Reuters reported on October 18, citing the Telegraph. Microsoft declined to comment, and that report did not establish its reason for leaving.
The original ambition was substantial. In an announcement dated October 15, 2025, Nscale said the campus would house approximately 23,000 Nvidia GB300 GPUs for Microsoft Azure services from the first quarter of 2027. It described an initial 50-megawatt facility, scalable to 90 megawatts. Those were deployment plans, not evidence that the capacity was operating.
The Loughton data center’s missing connection
Reuters also reported that grid officials had warned Nscale that power was not expected until the 2030s. Nscale says it remains committed to Loughton, is exploring alternative power sources and that its other Microsoft contracts are unaffected.
That distinction matters. Microsoft ending this agreement is not the same as Nscale abandoning the campus, or the two companies ending their wider relationship. An alternative electricity supply remains something to investigate, not a replacement already demonstrated to meet the site’s needs.
A queue is not computing capacity
The broader constraint predates this weekend. On July 29, 2026, Britain’s energy regulator Ofgem launched a consultation on data-center connection fees and stronger project milestones. It said contracted demand connection offers had risen from 41 gigawatts in November 2024 to 125 gigawatts in June 2025, largely driven by data centers.
Those figures describe requests and offers in a connection queue—not electricity already consumed, or a reliable forecast that every proposed facility will open. Ofgem’s concern was that projects unlikely to proceed could reserve scarce capacity and delay viable developments. Its July proposals should not be mistaken for a new rule announced today.
NeonPulse’s earlier coverage of AI infrastructure financing examined the cost of raising the money. Loughton brings a different delivery risk into view: a project can have a major customer and a processor plan without a matching electricity timetable.
The lesson is not that demand for AI has disappeared. It is that contracted ambition and usable capacity are different things. The next convincing milestone is not another chip count; it is a credible route to switching the machines on.
