Microsoft has withdrawn from its planned Loughton data-center agreement with Nscale, according to Reuters reporting carried by Euronext on October 18, citing the Telegraph. The Guardian also reports the withdrawal and quotes Nscale saying it remains committed to Loughton and is exploring alternative power sources. Those reports establish a new customer development, while the underlying agreements and termination terms are not available in the material reviewed here.

The distinction matters because our earlier coverage concerned the site’s grid delay. Repeating that delay would add little. A reported customer exit raises a separate question: which commercial commitment remains attached to the proposed capacity? It does not, on its own, establish that the project is cancelled, that every contract between the companies has ended or that demand for AI compute generally has fallen.

Separate the site from the customer agreement

Our analysis treats an infrastructure project as several records that need reconciliation. One describes the physical site and its power path. Another describes the service a customer has agreed to buy. A third describes when a workload can actually start using that service. An announcement can refer to all three without making their status identical.

Imagine a hypothetical facility whose initial customer needs capacity by a particular date. If the power timetable moves beyond that date, the customer might seek another location even while the developer continues the project. Conversely, a customer could remain committed while the physical delivery timetable is uncertain. These examples illustrate why contract status and site status need separate verification; they are not claims about the undisclosed terms of the Loughton agreement.

For a buyer evaluating a capacity proposal, the first useful question is therefore specific: which site, service period and customer commitment does this document cover? A statement about a company-wide relationship does not answer the status of an individual location. Nor does a change at one location reveal the status of another. Read the scope before turning a project-level update into a claim about an entire supplier.

A replacement power plan is a proposal until its dependencies are resolved

Exploring an alternative is different from establishing an operating supply. A project review would ask what evidence supports each step: the proposed source, the usable capacity, the expected availability and the conditions that still need to be satisfied. This is a proposed diligence framework, not confirmation that any particular option has been approved or selected for this site.

Keep the dates attached to those steps. A construction milestone, a power milestone and a customer service milestone can each be accurate while referring to different stages. A useful internal schedule would show the dependencies between them and identify which dates are commitments versus estimates. Without that distinction, a single opening date can conceal the condition on which it depends.

For an AI team, translate the proposal into a workload question. What would have to be true before a training run or production service could begin? Ask how a delay would affect the team’s existing plan, and identify a fallback before treating proposed capacity as available capacity. A hardware count or an artist’s impression cannot answer that operational question.

What would resolve the uncertainty

The strongest follow-up would be a current statement identifying the status and scope of the customer agreement, alongside a documented delivery plan for the site. If a replacement customer or power arrangement is announced, assess that new evidence on its own terms. Do not infer a completed replacement merely because the developer continues to express commitment.

There is also a boundary around financial interpretation. The reporting reviewed here does not supply the contract’s termination payment, replacement revenue or the cost of any alternative power plan. Our analysis therefore makes no estimate of a company loss, valuation effect or financing outcome. Unknown amounts should remain unknown rather than being filled with a plausible-looking calculation.

The new report is consequential because customer commitment and physical delivery are now both questions for this project. The practical response is to reconcile those records, not collapse them into a single verdict about whether AI infrastructure is viable.

Before relying on proposed compute capacity, verify the named site’s customer agreement, power milestone and service start date separately, with an owner for each unresolved dependency.