Microsoft reportedly exits UK AI data center agreement with Nscale amid power supply delays
Microsoft has reportedly withdrawn from its agreement to lease artificial intelligence computing capacity at Nscale's planned data center in Loughton, Essex, disrupting a project originally designed to deliver 23,040 NVIDIA GB300 GPUs for Microsoft Azure in the first quarter of 2027.
The reported withdrawal, which emerged on October 18, 2026, follows significant difficulties securing sufficient electricity for the facility. Grid access may not become available until the 2030s, creating a substantial discrepancy between the original deployment schedule and the site's potential operational readiness.
Microsoft has not publicly explained the decision, and the precise contractual reason for its withdrawal has not been established. Nscale maintains that it remains committed to the Loughton development and that its other international agreements with Microsoft are unaffected.
The case raises questions about the financial value of AI infrastructure contracts when customer commitments, hardware procurement and data center construction depend on electricity connections that have not yet been secured.
··········
THE ORIGINAL MICROSOFT–NSCALE AGREEMENT AND ITS TECHNICAL SCOPE.
The Loughton project combined a 50 MW AI computing facility, a large NVIDIA Blackwell deployment and a delivery target dependent on electrical infrastructure.
Nscale announced the acquisition of its Loughton site in January 2025, identifying the location as capable of supporting an initial 50 MW of AI and high-performance computing capacity, expandable to 90 MW. The company originally anticipated bringing the site into operation during the fourth quarter of 2026.
On September 16, 2025, Nscale announced its agreement with Microsoft to develop what it described as the United Kingdom's largest AI supercomputer. The planned facility would initially accommodate 23,040 NVIDIA GB300 GPUs, with deployment targeted for the first quarter of 2027.
The announcement formed part of a wider UK infrastructure initiative involving Microsoft, NVIDIA and OpenAI, with Nscale outlining plans for up to 58,640 GPUs across multiple British projects. That larger figure included separate initiatives and should not be attributed entirely to Loughton.
The original specifications distinguish planned infrastructure from delivered capacity.
........
Parameter | Announced specification |
Location | Loughton, Essex, UK |
Infrastructure provider | Nscale |
Intended customer | Microsoft Azure |
Initial electrical capacity | 50 MW |
Potential expansion | 90 MW |
Planned accelerators | 23,040 NVIDIA GB300 GPUs |
GPU deployment target | Q1 2027 |
Original site opening target | Q4 2026 |
Commercial development | Microsoft withdrawal reported October 18, 2026 |
........
The difference between the original site opening target and the subsequent GPU deployment target matters because constructing a data center does not mean its full computing capacity is ready for commercial use. GPU clusters require commissioned electrical distribution, cooling, high-speed networking and validated operational procedures before customer workloads can run reliably.
The GB300 platform was designed for high-density AI computing, making power delivery and liquid cooling central to the facility's engineering requirements. Neither the 50 MW capacity nor the planned GPU count demonstrates that the infrastructure was operational in October 2026.
··········
WHY ELECTRICITY AVAILABILITY THREATENS THE PROJECT'S ECONOMICS.
The reported grid-connection delay creates a potential mismatch between Microsoft's computing requirements and the date Nscale could deliver usable capacity.
The Loughton development has encountered substantial obstacles in obtaining electricity for its planned computing infrastructure. Information reported on October 18 indicates that a conventional grid connection may not be available until the 2030s, several years beyond the original GPU deployment target.
Nscale has acknowledged electricity-grid difficulties affecting the London area and indicated that it is exploring alternative power sources. However, the existence of these difficulties does not establish the contractual mechanism behind Microsoft's reported withdrawal. The company has not disclosed whether the decision resulted from an unmet delivery milestone, a termination provision, a negotiated commercial arrangement or another consideration.
A facility drawing 50 MW continuously would consume approximately 438 GWh annually, calculated from 8,760 operating hours. At 90 MW, the equivalent full-load figure would be 788.4 GWh. These are theoretical calculations based on continuous operation at rated capacity, not measured consumption or Nscale forecasts.
Electricity is required for GPUs and supporting processors, memory, networking, cooling and power conversion. Consequently, energy infrastructure must be engineered around the complete facility rather than the nominal power consumption of individual accelerators.
A substantial grid delay can postpone customer revenue while construction and financing expenses continue. It can also complicate hardware procurement: accelerators acquired before commissioning may remain idle, whereas deferred purchases introduce exposure to changes in pricing, availability and product generations.
Cloud providers plan computing capacity across multiple regions and suppliers, so extended delivery delays can make alternative facilities commercially preferable even when the original project remains technically feasible.
Alternative power arrangements would need to provide reliable electricity at the required scale. Operating at 50 MW for one day requires 1,200 MWh of energy; a battery capable of supplying that load for 24 hours would need at least 1.2 GWh of usable storage before conversion losses and operational reserves. Storage can support resilience but does not independently replace dependable long-term generation or grid supply.
The practical question is whether Nscale can secure a power solution with a credible commissioning date and commercially acceptable operating costs.
··········
THE INTERNATIONAL PARTNERSHIP AND NSCALE'S COMMERCIAL EXPOSURE.
The reported loss of Loughton concerns one facility within a larger Microsoft–Nscale program, making project-level contractual exposure more relevant than aggregate partnership value.
On October 15, 2025, Nscale announced an expanded agreement with Microsoft covering approximately 200,000 NVIDIA GB300 GPUs across Europe and the United States. It included planned deployments in Texas, Portugal and Norway, while reaffirming Loughton.
These projects have separate locations, infrastructure arrangements and deployment schedules. A setback at the UK site does not establish that the other agreements have been terminated.
........
Project | Announced infrastructure | Status relative to Loughton |
Loughton, UK | 23,040 GB300; 50 MW initially | Microsoft participation reportedly terminated |
Texas, US | Approximately 104,000 GB300; 240 MW initial campus | Separate US deployment |
Sines, Portugal | Approximately 12,600 GB300 | Separate European agreement |
Narvik, Norway | Approximately 52,000 GB300 | Separate Nscale–Aker deployment |
........
These figures describe announced deployments rather than verified GPUs operating simultaneously on October 18, 2026.
In September 2025, Nscale announced a $1.1 billion Series B financing round to support international expansion. Its infrastructure model requires substantial capital expenditure before computing capacity becomes operational and generates revenue.
Long-term hyperscaler agreements can support financing by establishing future demand, but their value depends on delivery obligations, contract conditions and commissioning. Reported discussions surrounding a potential US listing have mentioned a valuation of approximately $35 billion and a contracted portfolio exceeding $100 billion; these reported amounts are not audited revenue, cash receipts or guaranteed earnings.
The financial impact of the Loughton withdrawal cannot currently be quantified because the specific contract value, termination provisions and potential compensation have not been publicly established. Nscale could seek another customer, revise the timetable or restructure financing, but none of these outcomes is a confirmed corporate decision.
For investors, contracted capacity, physically installed GPUs, operational electrical capacity and revenue-generating services represent different development stages and different execution risks.
··········
WHAT THE LOUGHTON WITHDRAWAL MEANS FOR UK AI INFRASTRUCTURE.
The site's future depends on securing electricity and restoring a viable deployment schedule, while the reported customer withdrawal remains distinct from cancellation of the facility.
Nscale has stated that it remains committed to developing Loughton and is exploring alternative electricity supplies alongside other possible UK locations. That position leaves the project's future open, although the original Microsoft deployment cannot be assumed to remain commercially valid.
The reported grid constraints illustrate a broader UK infrastructure challenge. Large GPU installations require electrical capacity, cooling and network connectivity within the same timetable, while transmission upgrades and grid connections can take years. Accelerator generations and customer requirements may change substantially during such delays.
A facility originally designed around NVIDIA GB300 systems may therefore require technical and commercial reassessment if electricity becomes available several years later than planned.
The outstanding information concerns the contractual status of Microsoft's agreement, Nscale's revised power arrangements and whether the facility can secure a replacement customer on economically viable terms. Until those details are established, the evidence supports treating the development as a reported withdrawal from a specific computing-capacity agreement, not the end of the wider partnership.
For developers and investors, the appropriate measure of progress is deliverable computing capacity supported by confirmed electricity availability, an executable deployment schedule and enforceable customer commitments, rather than announced GPU counts alone.
··········
FOLLOW US FOR MORE.
·····
DATA STUDIOS
·····
