Microsoft’s plan to anchor a major UK AI data center has run into a problem that money and GPUs cannot quickly solve: grid access. The company has terminated its agreement for Nscale’s planned Loughton, Essex, facility after expected power availability slipped into the 2030s.
The setback puts a spotlight on a growing infrastructure challenge for the UK and wider EMEA market. The 50MW campus was expected to begin operating in early 2027 with 23,040 Nvidia GB300 GPUs supporting Microsoft Azure services in Britain.
For partners selling cloud, data center, and AI infrastructure, the question is increasingly whether new capacity can be powered on the timelines customers expect.
UK grid delays push Microsoft off the Loughton timetable
Reuters reported that electricity grid officials do not expect to supply the Loughton site until the 2030s. Grid connection bottlenecks are delaying some UK data center projects by as long as 13 years, creating a major obstacle to the country’s efforts to expand AI infrastructure.
Microsoft said its decision applies to a single site and that its wider partnership with Nscale remains intact.
Nscale is still pursuing the £2 billion Loughton development. The company told the BBC it is exploring alternative power sources and other UK expansion sites, while its Microsoft projects in Norway, Portugal, and the United States remain unaffected. The Essex facility is expected to create 250 permanent jobs if it eventually opens.
For EMEA partners, the Loughton setback illustrates how grid availability can influence where AI infrastructure is built, alongside connectivity, customer demand, and access to hardware. Projects in other regions may become more attractive when local power constraints threaten delivery timelines.
Neocloud economics get tougher when capacity arrives late
The cancellation also highlights the pressure facing neocloud providers building GPU-heavy infrastructure around a small number of large customers.
According to the Data Center Knowledge, Loughton could eventually scale from 50MW to 90MW. Neoclouds can be particularly exposed to delays because they often have concentrated customer bases, higher capital costs, and equipment plans tied to specific deployments.
“Capacity that isn’t energized generates no revenue, but the costs keep running,” Alex Cordovil, research director at Dell’Oro Group, told the publication.
Nscale reported $103.4 billion in active and contracted total contract value as of Aug. 31, although only $2.6 billion was classified as active. The gap illustrates how much of its reported business depends on projects and services that have yet to become operational.
Its broader Microsoft agreements could generate up to $43.8 billion through 2033, subject to delivery and service-availability requirements. Those figures should not be interpreted as the value of the canceled Loughton arrangement alone.
Power delays can also affect GPU economics. A site that opens years later than planned has less time to earn returns on expensive accelerators before newer generations change pricing and performance expectations.
What UK and EMEA channel partners should check before backing AI capacity
For UK MSPs, resellers, cloud partners, and infrastructure providers, Loughton shows why planned AI capacity should not automatically be treated as available capacity. Partners building customer projects around new UK facilities may need to account for longer grid timelines and the possibility that workloads will have to move elsewhere.
Partners should check:
- UK power and delivery dates: Confirm when facilities are expected to receive grid capacity and whether alternative generation is realistic. A multi-year connection delay can push customer deployments back even when hardware and contracts are already in place.
- Regional alternatives: Determine whether providers can move workloads or future builds to other UK or EMEA locations when one market cannot supply power on schedule.
- Contracts and GPU timing: Review delivery milestones, termination rights, and when accelerators will actually be installed, particularly when financing depends on specific customers or deployment dates.
For partners serving customers across the UK and Europe, the key issue is where usable AI capacity will actually come online. GPU counts and signed contracts matter, but grid access, site readiness, and backup locations may ultimately determine which providers can deliver when customers need the compute.
Also read: As AI data centers demand more power, partners are having to rethink infrastructure design, power delivery, and deployment planning.