AI News

Microsoft Walked Away From the UK's 'Largest' AI Data Center. The Problem Was a Power Line

Aditya Kumar JhaAditya Kumar JhaLinkedInAmazon·October 19, 2026·8 min read

Microsoft exited Nscale's Loughton AI data center after the grid delay slipped to the 2030s. What happened and what it says about AI power limits.

The UK's most heavily advertised AI data center just lost its anchor customer, and not because of chips, money or demand. On Monday, October 19, 2026, Microsoft confirmed it had exited a deal with British start-up Nscale to lease computing capacity at a planned facility in Loughton, Essex, northeast of London. The reason reported across several outlets is mundane and enormous at once: the site's grid connection slipped from the end of 2026 to the 2030s. (Source: Kuwait Times; Silicon UK)

If you only follow model launches, this looks like a footnote. It is actually a clean example of the constraint that increasingly decides how fast AI can grow, which is electricity and the wires that deliver it.

Insight

Quick summary: Microsoft confirmed on October 19, 2026 that it exited a deal to lease capacity at Nscale's Loughton, Essex AI data center. UK Power Networks reportedly told Nscale the site's power connection would slip from the end of 2026 to the mid-2030s. Microsoft says the decision "relates to a single site" and that it remains committed to its Nscale partnership; Nscale says it remains committed to Loughton. The episode comes as reports count at least 47 large US data center projects delayed, withdrawn or blocked in 2026, and as Nscale prepares a US IPO.

What happened at Loughton

Microsoft had agreed the previous year to lease computing capacity at a planned Nscale facility in Loughton. The deal was announced during President Donald Trump's state visit to the UK alongside multi-billion-dollar US tech investments, and Keir Starmer called Loughton the "largest UK sovereign AI data centre." (Source: Kuwait Times (AFP))

According to Silicon UK, earlier in October UK Power Networks, which runs the grid across London and the South-East, told Nscale the site's power connection would slip from the end of 2026 to the mid-2030s. Microsoft had been the only publicly announced client and expected anchor tenant. A spokesperson said the decision "relates to a single site" and that Microsoft remains "committed to our partnership with Nscale." (Source: Silicon UK; Kuwait Times)

Nscale says it remains committed to Loughton, is exploring other power sources, and calls its relationship with Microsoft "strong and long-term." The two continue to work together on projects in Norway, Portugal and the US. One source familiar with the matter told the wire report that Loughton was a small part of Microsoft's agreements with Nscale. (Source: Kuwait Times; Silicon UK)

The size of what was planned

A roundup citing the Telegraph and Reuters describes the site as planned at 50 megawatts, scalable to 90, with about 23,000 Nvidia GB300 GPUs for Microsoft Azure from early 2027. Network World, citing the Telegraph, reports the same figures (23,040 chips), so treat them as reported, not company-confirmed. (Source: Digg roundup)

Why a power line can stop an AI project

AI data centers draw power on the scale of small cities, and they need a grid connection before a single rack can run at full load. Silicon UK reported that in March 2026 the Loughton site was still operating as a scaffolding yard, which shows how early in construction the project was when the grid news arrived. UK Power Networks said many data centers are waiting on upgrades to the national transmission network, which it does not own. (Source: Silicon UK; Kuwait Times)

The wire report adds that some UK data center projects have stalled over high electricity prices and long grid-connection waits, and that OpenAI earlier this year withdrew from a planned UK "Stargate" partnership with Nscale and Nvidia, citing high energy costs and regulatory uncertainty. (Source: Kuwait Times)

SignalReported figureWho reported it
Loughton grid connectionSlipped from end of 2026 to mid-2030sSilicon UK
Planned site size50 MW, scalable to 90 MW; about 23,000 GB300 GPUsNetwork World, citing the Telegraph
US data center projects delayed or blocked in 2026At least 47RegLayer data cited by the FT, via Digg
Share of planned US data center pipeline Bernstein considers credibleRoughly 35%Bernstein, via the FT, via Digg
Nscale IPO ambitionNYSE listing, valuation up to $35 billion (reported)Kuwait Times wire report

Why investors are watching

Nscale is preparing a New York Stock Exchange IPO, reportedly targeting a valuation of up to $35 billion. Its filing says a large share of committed revenue comes from a few customers, namely Microsoft and Anthropic. Losing a named anchor tenant at a flagship site is therefore more than a local story. (Source: Kuwait Times)

Wider data points are worse for the sector. Reports summarized by Digg cite the Financial Times count of at least 47 large US data center projects delayed, withdrawn or blocked in 2026, based on RegLayer data the FT described as conservative, and Bernstein's view that only roughly 35% of the planned US data center pipeline is credible. Both figures are attributed to the Financial Times and reach us through summaries, so verify before you quote them. (Source: Digg summary; Digg roundup)

What it means for people who use AI

  • Compute is finite and local. Model quality gets the headlines, but where and when capacity can be switched on decides who can serve millions of users.
  • Cheaper tokens are not guaranteed. If capacity arrives later than planned, price cuts and usage-limit increases can slow. That is an inference, not something these reports state.
  • Single sites are fragile. Microsoft said this decision concerned one site, and the partnership continues elsewhere, which is exactly why large buyers spread capacity across regions.
  • Do not over-read one deal. Microsoft and Nscale still work together in Norway, Portugal and the US, and the Loughton exit does not by itself show AI demand falling.
Pro Tip

Use this as a prompt to avoid lock-in on the user side too. If your workflow depends on a single model or provider, test one or two alternatives now, while nothing is broken, so a capacity crunch or a policy change does not force a rushed switch later.

Frequently Asked Questions
01What exactly did Microsoft do?

It confirmed on October 19, 2026 that it exited a deal to lease computing capacity at Nscale's planned Loughton, Essex data center. A spokesperson said the decision relates to a single site.

02Why did the project stall?

Reports say UK Power Networks told Nscale the site's grid connection would slip from the end of 2026 to the mid-2030s. UK Power Networks also said many data centers are waiting on national transmission upgrades.

03Is Microsoft leaving Nscale entirely?

No. Microsoft says it remains committed to its partnership with Nscale, and the companies continue to work on projects in Norway, Portugal and the US.

04Is Nscale still going public?

Nscale is reportedly preparing a New York Stock Exchange IPO at a valuation of up to $35 billion. Timing was unclear in the coverage I reviewed.

05Does this mean the AI data center boom is over?

Not on this evidence. The reports describe delays and cancellations, but a single site exit does not show demand collapsing. The 47-project and 35% figures are secondary and worth verifying.

Infrastructure stories like this one are a reminder that AI is a physical industry with real limits. When you want answers without worrying which data center they come from, LumiChats gives you 40+ premium AI models side by side in one subscription, so you can pick the model that works best for each task.

Was this article helpful?

Found this useful? Share it with someone who needs it.

Free to get started

Claude, GPT-5.4, Gemini —
all in one place.

Switch between 40+ AI models in a single conversation. No juggling tabs, no separate subscriptions. Pay only for what you use.

Start for free No credit card needed
Aditya Kumar Jha
Written by
Aditya Kumar JhaLinkedIn

Published author of six books and founder of LumiChats. Writes about AI tools, model comparisons, and how AI is reshaping work and education.

Keep reading

More guides for AI-powered students.