Summary
- Nscale’s planned Loughton AI campus is reportedly affected by delays to a 90MW-scale grid connection.
- The site was previously announced as a Microsoft-backed AI campus with 50MW of capacity, scalable to 90MW.
- The case exposes the gap between AI data centre build timelines and UK electricity connection delivery.
Nscale is reportedly exploring alternative power arrangements for its planned AI campus in Loughton, Essex, after delays to the site’s grid connection disrupted the project’s power timetable.
The campus was announced in 2025 as a UK AI infrastructure project backed by commitments involving Microsoft, Nvidia, and OpenAI. Nscale said at the time that the Loughton site would deliver 50MW of AI capacity, scalable to 90MW, and would initially house 23,040 Nvidia GB300 GPUs from the first quarter of 2027 to support Microsoft Azure services in the UK.
Recent reporting says the grid connection will not be ready in time, with the company examining bridge power or alternative supply options. Bloom Energy fuel cells have been cited as one possible route. Nscale has not published a new detailed power plan for the site.
The queue meets the GPU cycle
The Loughton case gives a practical shape to the UK’s data centre power problem. AI campuses can be marketed, financed, and built faster than the electricity network can always connect them. That leaves developers trying to match customer demand and hardware deployment cycles with a grid process measured in years.
AI facilities intensify the mismatch because their power profile is large, dense, and difficult to phase gently. Conventional enterprise or colocation capacity may grow through halls and customer deployments over time. A GPU-heavy AI facility can need substantial power early in its operating life because the value of the site depends on bringing dense compute online while the hardware is commercially current.
Behind-the-meter power is becoming a more common response. Developers are looking at fuel cells, gas engines, batteries, private wire renewables, and hybrid systems to start operations before a full grid connection arrives or to reduce exposure to network constraints. Those approaches carry trade-offs. Fuel cells and gas generation can provide firm power, but they raise emissions, fuel supply, planning, noise, permitting, and maintenance questions. Solar and batteries can improve the mix, but they do not easily provide continuous baseload for a 90MW AI campus without major land and storage requirements.
Power quality matters as much as the source. AI workloads need stable electricity, redundancy, protection, maintenance access, and integration with UPS and backup systems. A bridge-power arrangement that gets a site live but introduces fragility would move risk from the grid queue into the facility.
AI policy still needs substations
The UK government has placed AI infrastructure high on the national agenda and has designated data centres as critical national infrastructure. It has also moved to accelerate strategically important data centre projects and reform the electricity connections process. Those signals do not remove the practical lead times attached to substations, transformers, switchgear, grid reinforcement, protection systems, and high-voltage equipment.
Large data centre loads are competing with renewables, battery projects, housing growth, electrified transport, heat decarbonisation, and industrial electrification for connection capacity and network investment. That competition has exposed the weakness of a system built around long queues rather than the readiness, strategic value, and deliverability of individual projects.
For Nscale, delays are commercially awkward. AI infrastructure customers want capacity inside current hardware and service cycles. GPU deployments depreciate financially and technologically. A late connection can turn a strategic site into an expensive holding pattern, especially where customer commitments depend on capacity being available by a specific date.
The issue also changes the local planning profile of AI campuses. If bridge power becomes common, a data centre project can start to look like an energy project as well as a digital infrastructure project. On-site generation, fuel logistics, emissions modelling, acoustic design, heat rejection, fire safety, and environmental permitting all become part of the same development package.
The UK does not lack AI capacity ambition. The harder task is turning demand into powered, permitted, operational facilities on timelines that match the market. Loughton shows the constraint plainly: the compute can be sold, the GPUs can be specified, and the building can be planned, but the power connection still sets the pace.
