Microsoft pulls out of AI datacentre project in Essex over power supply concerns

Microsoft pulled out of its planned customer agreement for Nscale’s Loughton datacentre in Essex. It was expected to be the anchor tenant, meaning the main customer helping justify and support the facility. Losing that commitment weakens the project’s commercial foundation.

The withdrawal follows a serious infrastructure problem. UK Power Networks told Nscale that it could not supply enough electricity to the site until the early to mid-2030s. Without sufficient power, the datacentre cannot begin operating as planned next year. The location had also drawn scrutiny because it was still an operational scaffolding yard when the Guardian reported on it in March.

Nscale says it remains committed and is exploring onsite generation and faster grid connections. However, Microsoft’s departure leaves no publicly announced customer for the Essex facility. That could make it harder to attract funding, new tenants, and investor confidence ahead of Nscale’s proposed Wall Street listing.

An AI datacentre is a large facility filled with computers, networking equipment, and supporting systems for artificial intelligence. It provides the processing capacity needed to train AI models and operate them for users. The article describes datacentres as the central nervous system of technologies such as chatbots.

Training involves processing enormous collections of data to identify patterns and adjust a model. Once trained, the model still needs powerful computers to answer many users at once. Those computers work continuously and require cooling, networking, and reliable electricity. This makes power availability as important as the buildings themselves.

The Loughton project was promoted as the UK’s largest sovereign AI datacentre. Its planned delay shows the practical challenge behind AI expansion: ambitious software services need physical infrastructure. If electricity or connections are unavailable, even a well-funded project cannot quickly deliver the computing capacity that AI companies expect.

The article does not state how much electricity AI datacentres typically use. Established industry estimates put smaller AI facilities in the tens of megawatts, while large hyperscale sites can require 100 megawatts or more. A megawatt measures the rate of electricity use, so these are substantial continuous demands.

For comparison, an ordinary large office building usually needs only a few megawatts or less. A major hospital, factory, or transport facility may use several to tens of megawatts, depending on its equipment and operation. The largest AI sites can therefore resemble heavy industrial users or the demand of a small town rather than a normal commercial building.

That scale helps explain Loughton’s difficulty. UK Power Networks said it could not supply enough electricity until the early to mid-2030s. The exact capacity requested by Nscale is not given, so no precise comparison can be made. The article’s central point is that energy has become a major constraint on datacentre development.

The immediate problem is a shortage of available grid supply for Nscale’s planned datacentre. UK Power Networks, which owns and maintains the electricity network across London and south-east England, told Nscale that the site could not receive enough power until the early to mid-2030s. That timetable is far beyond the project’s planned opening next year.

The article does not explain whether the delay comes from local capacity, network upgrades, connection queues, or another technical constraint. It does say the issue is London-wide and that the failure to find a quick fix shows how energy has become a hindrance to datacentre development. The project therefore depends on infrastructure that cannot currently be delivered quickly.

Nscale is investigating whether it can generate electricity onsite and speed up the connection process. Until one of those options works, the facility may remain delayed. The problem also threatens the government’s wider ambition to expand sovereign AI infrastructure in Britain.

Microsoft was expected to be the anchor tenant, so its departure removes the project’s main publicly known commercial commitment. An anchor tenant can help a datacentre secure financing, attract other customers, and demonstrate that its planned capacity has real demand. Nscale now has no publicly announced customer for Loughton.

The loss comes alongside the power delay. UK Power Networks cannot provide enough electricity until the early to mid-2030s, while the facility was planned to open next year. The combination of an uncertain opening date and no announced customer makes it harder for Nscale to show that the site can generate revenue soon.

The article says this could undermine Nscale’s ambition to float on Wall Street this year at a potential valuation of $35bn. It also highlights the broader financial vulnerability of neoclouds, which are taking on increasingly risky debt while delays accumulate. Microsoft remains involved in Nscale’s Norway project.

Nscale’s clearest alternative is to produce electricity at the Loughton site rather than depend entirely on the delayed grid connection. The company says it is investigating whether onsite generation is possible. This could address the immediate power shortage, although the article does not say what technology would be used or whether it would provide enough capacity.

The company is also exploring whether the connection process can be accelerated. That means seeking a faster route to the electricity network than the current early-to-mid-2030s timetable. The article does not specify which upgrades, agreements, or technical changes might achieve that. It only confirms that Nscale is examining the possibility.

Nscale is also considering other expansion sites in the UK. Its spokesperson says the company remains committed to Loughton and the UK. However, until an alternative power source or quicker connection is secured, the site remains exposed to the same electricity constraint that has already delayed the project.

AI systems learn by processing large datasets and repeatedly adjusting mathematical parameters to recognise patterns. This training requires many computing operations carried out across powerful processors. The process can run for long periods, creating heavy electricity demand. The article does not explain these technical details, but it identifies datacentres as the central nervous system of AI technology.

After training, an AI model still needs computing power whenever someone uses it. A chatbot must process a request, apply the model, and produce an answer. Many simultaneous users multiply that workload. Datacentres provide the specialised hardware, networking, storage, cooling, and reliable power needed to keep these services running.

That dependence explains why the Loughton power problem matters beyond one building. The government wants AI to support economic growth and sovereign capability, while companies such as Nscale and CoreWeave build the infrastructure. But if electricity connections are delayed, expansion of AI services is limited regardless of demand or investment.