Getty Images/iStockphoto
Microsoft's Nscale exit exposes $103B AI contract pipeline risk
Microsoft terminated its Nscale data center contract after UK power delays. Nscale's $103B contract pipeline highlights the financing risks facing AI infrastructure developers.
Microsoft terminated its contract for Nscale's planned AI data center in Essex, England, after a power-delivery delay threatened the project's timeline.
The setback highlights a risk for data center developers and the operators buying AI capacity: Customer contracts and financing commitments cannot generate revenue until facilities have power and can deliver computing capacity.
For developers, delays can leave construction and financing costs running while customer payments remain tied to delivery milestones. For operators, a missed power deadline can leave planned AI capacity unavailable when needed. Neocloud providers face particular exposure because they often depend on a handful of large customers and borrow against specific deployments.
Nscale unveiled the Loughton project in 2025 as a 50-megawatt campus scalable to 90MW, initially equipped with 23,040 Nvidia GB300 GPUs for Microsoft Azure. The company confirmed through a spokesperson that Microsoft terminated the contract. Nscale remains committed to the site and is exploring alternative power sources and other UK expansion sites. The spokesperson said the broader relationship with Microsoft remains strong and the Loughton issue does not affect other contracts.
The Guardian reported that UK Power Networks had told Nscale it could not supply enough electricity for the site until the early to mid-2030s, years after the project's planned opening. The companies have not disclosed the financial consequences of the termination.
The termination puts a spotlight on the scale of Nscale's contracted business and the conditions required to spark revenue. Its September IPO filing reported $103.4 billion in active and contracted total contract value as of Aug. 31, 2026, compared with $2.6 billion in active contract value.
Microsoft agreements provided for payments of up to $43.8 billion through December 2033, subject to delivery and service-availability requirements. Separate agreements with Anthropic provided for up to $44.6 billion, also subject to delivery requirements. Nscale said it had not obtained binding financing commitments for the GPU equipment and infrastructure needed to fulfill the Anthropic agreements as of the prospectus date.
"Capacity that isn't energized generates no revenue, but the costs keep running," said Alex Cordovil, research director at Dell'Oro Group.
Power is the missing link
AI infrastructure projects depend on three elements lining up: a creditworthy customer, financing supported by that commitment, and available electricity, Cordovil said. Developers have more control over securing customers and financing than over when utilities can deliver power.
Grid operators sometimes revise connection commitments by six months or a year, and occasionally longer. Permitting that depends on third parties can also delay projects, he said.
"The problem isn't necessarily that AI demand disappears," said Neil Osnato, founder of Persistence Analytics Group. "It's that capital can become committed long before the physical and financial assumptions supporting its recovery are proven."
Neoclouds face concentrated exposure
A delayed site can have a greater financial impact on neoclouds than on hyperscalers, Cordovil said. Neoclouds often have concentrated customer bases, higher capital costs and GPUs financed against specific contracts.
"A hyperscaler can absorb one delayed site across a large, diversified fleet and balance sheet. A neocloud typically has more concentrated customers and a higher cost of capital," Cordovil said.
Losing an anchor customer can leave a provider with equipment and financing arranged around a deployment that no longer has a buyer. Nscale's filing describes multiple financing facilities for GPU equipment and data center infrastructure, but does not establish how much financing or equipment spending was associated with Loughton.
"The issue with neoclouds is that the debt and financing is colossal, and the demand they're betting on is not assured," said Owen Rogers, senior research director of cloud computing at Uptime Institute Intelligence.
Nscale has not publicly detailed the project's spending, termination payments or contractual terms. Those details would help establish how much of the project's financial risk falls on the company, its lenders or equipment suppliers.
Contract value isn't operating revenue
Nscale's filing describes long-term commercial arrangements, including take-or-pay contracts. Its Microsoft agreements remain subject to delivery and service-availability requirements, and contract terms generally commence after GPU clusters or infrastructure have been successfully delivered.
Cordovil recommended separating AI infrastructure announcements into three categories: announced, contracted and energized.
Commitments must also last long enough for providers to recover their investment, Rogers said.
Contracted capacity is meaningless until it's energized capacity,” he said. "Backlog only turns into cash if the neocloud can actually build and power the capacity on time.
Delays can erode GPU returns
Power delays can reduce the time developers have to earn returns on expensive accelerators before newer generations change the economics.
"Nvidia is now releasing new generations roughly every year, so today's best GPU won't be top-end for long," Rogers said. Older hardware can retain value for inference and less demanding applications, he added.
Providers can lose money even when their GPUs remain useful. Late deployment, lower utilization or revenue below financing assumptions can undermine returns, Osnato said. If a provider cannot cover its costs, it may have to sell hardware or the business at a discount.
Gartner has warned that substantial AI infrastructure investment could face impairment if projected revenue fails to support the capital committed. A TechTarget story published Oct. 14 examined that forecast and its implications for the market.
Cordovil cautioned against interpreting Loughton as evidence that AI infrastructure demand is weakening. The problem, he said, is that some project schedules assume grid connection dates utilities cannot meet. Behind-the-meter generation can help bridge the gap, but adds cost, permitting and execution risk.
Shane Snider is a senior news writer at TechTarget, covering AI infrastructure, hyperscale data centers, cloud platforms, and the power and energy systems driving modern compute expansion. You can reach Shane at [email protected] or on LinkedIn.
