Urgent.News

What's breaking now, across thousands of outlets.

AI

UK's biggest AI supercomputer may have to wait until 2030s for enough power

The UK start-up says the data centre could eventually expand from 50 to 90 megawatts of power — enough, if used continuously, to consume as much electricity in a year as about 315,000 typical UK homes, based on 2026 figures from UK energy regulator Ofgem.

The United Kingdom's largest planned AI supercomputer, set to launch next year, may face a delay until the 2030s, due to difficulties securing sufficient electricity. This project was announced as part of the first technology agreement between the UK and the United States, as the UK seeks to boost its artificial intelligence industry.

The site, being developed in Loughton, east of London, by Nscale, a UK-based firm that includes US chipmaker Nvidia as an investor, is expected to host approximately 23,000 Nvidia AI chips by 2025. These chips will be interconnected to form what Nscale claims will be the United Kingdom's largest AI supercomputer, intended to support Microsoft's Azure cloud services.

However, UK Power Networks (UKPN), responsible for connecting the site to the local power grid, has informed Nscale that the grid may not have enough electricity to supply the site until the early to mid-2030s. This concern stems from the need for upgrades to the broader national transmission network. National Grid, which manages England and Wales' power grid, is collaborating with NESO, the national system operator, and local network operators to expedite these connections.

Nscale is also exploring on-site power generation and discussing potential ways to expedite the connection with UK Power Networks.

The data centre at Loughton could eventually expand from 50 to 90 megawatts of power, which, if used continuously, would consume as much electricity annually as roughly 315,000 typical UK homes, according to figures from UK energy regulator Ofgem. This Loughton site is part of Nscale's broader plan to install nearly 59,000 Nvidia AI chips across data centres in the UK.

On a European scale, data center electricity consumption is projected to increase from 96 terawatt-hours in 2024 to 236 terawatt-hours by 2035, as highlighted by UK-based energy think tank Ember.

Recent developments have seen Scotland halt new approvals for hyperscale AI data centres due to concerns over whether the country's energy grid can handle the rapidly rising demand. Similarly, Amsterdam has restricted new data centres and expansions over issues of space and power, while a proposed bill in Denmark would grant large electricity users, including data centres, lower priority in grid connections.

Written by urgent.news from Euronews's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.

Read the original at euronews.com →

More in AI

I went to a Jev hackathon to find out why the new AI model is taking San Francisco by storm

TypeSafe's Jev has taken San Francisco developers by storm and could mark a shift in AI. At a hackathon, engineers rushed to build software with Jev.

  • Jev, a new AI model, is gaining attention in San Francisco's tech scene.
  • Created by TypeSafe AI, Jev differs from traditional LLMs by making decisions.
  • At a hackathon, Jev's speed and cost-effectiveness impressed developers and investors.

How to Translate Document Text With Local AI in 2026

You need to understand a document in another language, but do not want to upload it to a translation service. OGAD (Off Grid AI Desktop) can translate selected text with a model running on your…

  • Use supported Mac or Windows system with OGAD software
  • Download local text model for source and target languages
  • Translate one short passage at a time, verify details

How Much Does a Custom AI Document Assistant Cost?

A custom AI assistant that answers questions over your own documents typically takes a small team three to six weeks to build to a usable first version, and the cost is set mostly by how messy your…

  • Custom AI document assistant development takes 3-6 weeks for a usable first version.
  • Costs depend on document complexity, access restrictions, and data handling rules.
  • Deployment method (cloud API vs self-hosted) significantly impacts overall costs.

More from Tuesday 29 September →