Microsoft has withdrawn from Nscale's planned AI data centre in Loughton, Essex, while saying its broader partnership with the company continues. The development matters to organisations planning future UK compute capacity: the site that featured in an announced deployment is no longer backed by the same anchor-customer commitment.
In ITPro's October 19, 2026 report, Microsoft limited the decision to one site. Nscale said it remained committed to Loughton and was investigating alternative power sources. Both are company statements about scope and intentions, rather than evidence of a replacement power supply. Gizmodo also covered the withdrawal on October 19; the sources reviewed do not establish the exact date the cancellation took effect.
The original deployment was a future commitment
Nscale's September 16, 2025 announcement described an initial deployment of 23,040 Nvidia GB300 GPUs at Loughton, targeted for the first quarter of 2027. It presented 50MW of AI capacity, scalable to 90MW, intended to support Microsoft's Azure services in the UK. Those figures described the plan; they should not be counted as computing already available.
That same announcement treated Stargate UK as a separate initiative across multiple sites. Microsoft's Loughton decision therefore does not, by itself, establish the cancellation of every Nscale project or the entire UK programme.
Power delivery remains a material uncertainty. Data Center Dynamics reported on October 19 that the expected grid connection had slipped into the 2030s. It also reported National Grid's statement that it had not agreed the previously cited 2026 target. This is a reason to distinguish the parties and their milestones, not to substitute a new guaranteed opening date.
An AI service needs an accepted delivery milestone
For customers, the practical question is which facility can provide the required service on the required date. TechKili's assessment is to ask separately about the power connection, equipment commissioning and the point at which a contracted workload can actually begin. An equipment count does not answer all three.
A procurement discussion should also make clear what happens if those milestones move: whether a substitute site is offered, whether its location meets the customer's requirements, and how acceptance is documented. These are questions for the proposed service, not claims about undisclosed terms in the Microsoft agreement.
The evidence supports a site-specific setback and an unresolved delivery problem. It does not quantify the cancelled contract's value or establish the outcome of Nscale's wider financing plans. A revised, attributable commissioning schedule would be more useful to an AI customer than another aggregate GPU commitment.
Sources
Related Articles
RTX Spark brings CUDA PyTorch to Windows Arm64, with dependency gaps
Nvidia documents native Windows Arm64 builds as RTX Spark laptops arrive. Developers still need to match Python wheels, extensions and drivers to the complete workload.
VitVio's Vetro rollout in Gdańsk puts operating-room timelines first
The selected-room deployment targets surgical-stage logging and coordination. Its value depends on measured workflow results and clear handling of captured data.
Gemini 4 Argon opens wider, with different access rules for each product
The October 13 rollout brings Argon to paid API users and Gemini subscribers, but app access and coding access have different requirements.
OpenAI's Dublin Applied AI Lab will build with customers
The 200-job plan spans customer engineering, privacy, sales and operations. Its lab and Irish founder-training programme offer different forms of support.