1. Microsoft exits the Loughton project
Microsoft has withdrawn as anchor tenant of a planned AI data center in Loughton, Essex, in October 2026, after UK Power Networks told the developer, Nscale, that adequate grid power would not reach the site until the early to mid-2030s. The campus was meant to host 23,040 Nvidia GB300 chips for Microsoft's Azure cloud and open around the turn of the year.
The reason given for the exit is electricity, and nothing else. The reporting around the decision points to no shortage of chips, no funding gap and no change in Microsoft's appetite for AI capacity. The site simply cannot be supplied with the power it needs on any timetable that makes sense for a facility of this kind.
That distinction matters because it separates this episode from the usual story of an AI project stumbling over cost or demand. Here the customer, the developer and the hardware were all lined up. What failed was the connection between a building and the electricity network that was supposed to feed it.
2. What was planned for the Essex site
The facility was presented as the largest sovereign AI data center in the United Kingdom. In this context, sovereign means that the computing capacity sits inside the country and is operated under domestic control, so that sensitive workloads do not have to run on servers located abroad.
Nscale designed the first phase for 50 megawatts of capacity, with room to expand to 90 megawatts. A megawatt figure is the standard way to size a data center, because the electrical supply, rather than floor space, usually sets the limit on how much computing can be installed. A 50 megawatt site is large by any measure, and the planned expansion would have taken it considerably higher.
Inside, the plan called for 23,040 Nvidia GB300 graphics processors, all of them dedicated to Azure workloads. Committing an entire site to a single cloud customer is what makes that customer an anchor tenant: its long-term commitment is what allows the developer to finance construction and order equipment in the first place.
3. Why did the power connection derail the deal?
A data center that cannot be connected to the grid cannot run, and a delay measured in years is fatal for hardware that is expected to earn its keep quickly. UK Power Networks indicated that sufficient supply would not be available before the early to mid-2030s, a wait far longer than the few months that had been expected before opening.
For a tenant, the arithmetic is unforgiving. Advanced AI processors are bought for specific model-training and inference demand, and they lose value as newer generations arrive. Paying for, or even reserving, capacity that cannot be switched on for most of a decade offers no useful return, so withdrawing was the practical response.
Microsoft has described the move as a decision about one location rather than a verdict on the relationship. In other words, the company is saying that the grid timetable made this particular site unworkable, not that it has lost confidence in Nscale as a partner.
The case also shows how exposed a tenant is to factors outside the contract. A cloud provider can negotiate price, delivery dates and hardware specifications, but it cannot compel a network operator to energize a site sooner. When the operator's answer is the 2030s, the commercial agreement loses its footing, and the anchor tenant's only sensible choice is to look elsewhere for capacity.
4. How does a data center get its power?
A large data center does not simply plug into the nearest socket. The developer applies to the network operator for a connection of a stated size, and the operator must confirm that local cables, substations and transformers can carry that load. When they cannot, reinforcement work is required, and that work can take years to plan, approve and build.
AI facilities make the challenge sharper than conventional cloud buildings. Racks filled with advanced graphics processors draw far more electricity per square meter than traditional servers, and they run close to continuously. A campus of 50 megawatts is comparable to the demand of a small town, and one that could grow to 90 megawatts asks even more of the local network.
This is why site selection for AI capacity has shifted. Land, fiber and tax treatment still matter, but the first question developers now ask is where reliable power can be delivered, and how soon. A location that looks attractive on every other measure can be unusable if the connection date lies a decade ahead.
The pressure is not unique to Britain. Developers in many countries compete for the same limited network capacity, which is one reason on-site generation, whether from fuel cells or other equipment, has become a more common part of project plans. Such systems produce electricity next to the facility, allowing it to operate before, or without, a large grid connection.
5. What Microsoft and Nscale say about their wider relationship
Microsoft has said its broader partnership with Nscale remains strong. The company has separate agreements with Nscale covering projects in Norway, Portugal and the United States, and those contracts are not affected by the Loughton decision.
Nscale itself has not given up on Essex. It is looking at generating power on the site, with fuel cells among the options, which would reduce its dependence on a grid connection that is years away. The developer is also continuing to pursue other expansion opportunities in the United Kingdom.
On-site generation is a recognized workaround for power-constrained projects, but it brings its own questions about cost, fuel supply, permitting and reliability. Whether it can supply a campus designed around 50 to 90 megawatts of demand, and whether it can win back an anchor tenant, has not been settled.
6. Other Nscale sites are running behind as well
Loughton is not the only Nscale project facing delays. The company's flagship site in Norway is also behind schedule, with 6.4 megawatts of a planned 30 megawatts currently operational. That is roughly a fifth of the intended capacity at a location that was meant to demonstrate the developer's ability to deliver.
Two sites falling short in different countries suggest that the difficulty is not confined to one national grid. Building shells and installing servers can be done on commercial timelines. Securing power, transformers and network upgrades often cannot, and the developers who depend on them have limited control over the pace.
For customers weighing where to place long-term AI commitments, such delays raise the importance of checking power availability before signing, rather than assuming that it will be there when the hardware arrives.
7. A setback for Britain's AI ambitions
The cancellation is awkward for British policymakers because the project was closely tied to the government's goal of making the United Kingdom a leader in artificial intelligence. The ambition had been showcased at high-profile technology diplomacy events involving the United States and the United Kingdom in 2025, where large infrastructure commitments were a central theme.
A flagship sovereign facility that cannot be powered undercuts that message. Policy statements about attracting investment carry less weight when a major tenant leaves a headline project because the electricity network cannot deliver, and other prospective investors will be watching how quickly the problem is addressed.
The grid itself is the heart of the issue. The British network has a well-documented backlog of connection requests, and projects that need large amounts of power, including data centers, must queue behind others for upgrades. Faster connections would require investment and reform in the network, not just enthusiasm for AI.
8. What does this mean for Nscale's IPO plans?
Nscale is the party most directly affected. The company filed for a United States initial public offering in September 2026, and its prospects as a listed company rest mainly on its contracts with Microsoft and Anthropic. Losing Microsoft as anchor tenant at one site only weeks after that filing is poor timing.
Microsoft's statement that its other contracts remain in place helps contain the damage, since investors will want reassurance that the loss is limited to a single location. Yet the shortfall at the Norwegian site invites a further question: how many of Nscale's other projects face similar power constraints, and how firmly are their schedules grounded?
A company valued on capacity it intends to build must show that it can bring that capacity online. Contracts signed in advance are only worth what the developer can deliver, and power access is now one of the most important variables in that delivery. How Nscale handles Essex, through on-site generation or other solutions, will be an early test of its execution.
9. What to watch next
The immediate question is whether Nscale can secure an alternative power source for Loughton and attract a replacement tenant. Fuel cells and other on-site generation are under consideration, but no timetable or capacity figure has been announced for them.
Investors will also look for updates on the Norwegian site, where progress toward the planned 30 megawatts is behind schedule, and on the company's other UK expansion plans. Any further tenant changes or delays would carry greater weight given the pending listing.
For the wider market, the episode is a reminder that AI infrastructure is limited by physical supply as much as by chips and capital. Governments that want data centers within their borders will be judged on how quickly they can connect them to the grid.