Nscale's UK AI supercomputer may wait until the 2030s for power

Nscale's planned AI datacentre in Loughton, Essex, may not get enough grid power to operate until the early to mid-2030s, years after its expected 2027 debut. The delay puts founder Josh Payne's central wager - building AI infrastructure across power, data centres and computing - up against a problem that cannot be solved simply by ordering more GPUs.

The Guardian reported the delay on September 24th, citing information from UK Power Networks (UKPN), which is due to connect the site to the local grid. UKPN declined to comment on Nscale's project specifically. In general, the network operator said, many datacentres connected to distribution networks are waiting for work on the wider national transmission network before they can operate.

Nscale said it remains committed to the project and is examining on-site power generation and ways to accelerate its connection. Neither option yet establishes how much of the planned facility could run before a full grid connection becomes available.

Payne's infrastructure bet meets its first constraint

Payne's route into AI infrastructure ran through businesses in recruitment, capital markets and supercomputing, according to Nscale's leadership biography. Nscale grew out of Arkon Energy, an infrastructure business associated with bitcoin mining before its expansion into AI computing. That experience helps explain Payne's thesis: large-scale AI capacity needs power and physical facilities alongside GPUs and cloud software.

The Loughton project shows where that integrated model still depends on others. Nscale can develop the site and explore generating electricity behind the meter, but UKPN's account points to work on the wider transmission system as a gating factor. That leaves the project's schedule exposed to infrastructure beyond Nscale's control, even as Nscale markets power as part of its own full-stack offer.

When Nscale announced the Loughton site on January 13th, 2025, it said the facility would provide 50MW of AI and high-performance computing capacity, with room to scale to 90MW, and be live in the fourth quarter of 2026. The announcement said the site could house up to 45,000 Nvidia GB200 GPUs. The Guardian now describes a planned 2027 debut; either target is far earlier than the early-to-mid-2030s power estimate reported this week.

The site has already had a difficult path from announcement to construction. In March, the Guardian reported that the Loughton location was still being used as a scaffolding yard. Its latest report says Nscale has since cleared the site. That is a concrete step, but clearing land does not supply electricity or settle a commissioning date.

A crowded queue, and a public-market timetable

Loughton is seeking up to 90MW, which the Guardian compares with the electricity use of about 315,000 homes. The national queue is larger still: Ofgem said around 73GW of data-centre demand was represented in Britain's connections queue, and warned that speculative applications can hold up projects that are ready to proceed. The regulator's proposed reforms include commitment fees and project milestones intended to clear out stalled or non-viable applications. The queue figure describes requests, not power already allocated for use by operating datacentres.

Nscale's delay lands as the company prepares to ask public-market investors to assess a business built around deploying infrastructure at scale. Nscale filed an IPO registration statement with the U.S. Securities and Exchange Commission on September 18th. RuntimeWire's coverage of that filing examined the first-half financial figures Nscale reported. A late grid connection at Loughton does not determine the outcome of the IPO or of Nscale's other projects. It does make delivery timelines a harder part of the investment case to wave away.

Nscale has meaningful capital behind that expansion. In March, the company announced a $2 billion Series C, led by Aker ASA and 8090 Industries, and said the round valued Nscale at $14.6 billion. Those investors are backing a buildout whose pace depends partly on access to electricity networks, not just on financing, equipment orders or customer demand.

The UK government had cited Loughton in its 2025 AI strategy as the country's "largest UK sovereign AI datacentre." That label made the project a symbol of national computing ambitions; the connection delay exposes the less photogenic prerequisite underneath them. A data centre's advertised capacity becomes useful computing only when the site, power supply and network are ready together.

Nscale's response leaves two practical routes open: secure an earlier connection or generate power on site. An on-site solution could potentially reduce reliance on the local grid, but the information available so far does not establish its scale, timing or whether it would support the full project. Until one of those routes produces a workable schedule, the headline capacity remains a plan rather than operating compute.

For Payne, Loughton is therefore both a specific project and a test of the infrastructure thesis he has been building. His earlier work supplied experience in energy and compute-intensive sites; this project depends on converting that experience into an operating facility on the government's original timetable. Nscale says it remains committed. The date that matters now is when the site can draw enough power to run.