Nvidia in talks to back billions in debt for Nscale data center expansion
The chipmaker is discussing credit guarantees that could make it cheaper for the UK neocloud to borrow for its next big build
Nvidia is discussing billions of dollars in credit guarantees for Nscale. The guarantees would back the debt Nscale needs to build a 460-megawatt data center.
The world’s most valuable chip supplier already sells Nscale its GPUs, rents its capacity, and owns a piece of the company. Co-signing its loans would add yet another role to the relationship.
What’s on the table
The talks center on Nvidia standing behind Nscale’s borrowing for the 460-megawatt project. A credit guarantee works a lot like a parent co-signing a mortgage. If the borrower stumbles, lenders can turn to the deep-pocketed backer.
The 460-megawatt figure lines up with plans at Nscale’s Monarch Compute Campus in West Virginia. Nscale is negotiating an expansion there, with a second building planned for Anthropic.
That second building, which would have 460 MW of capacity, is still in discussions with Anthropic. It would use Nvidia’s Vera Rubin chips as part of multi-billion-dollar compute agreements.
This isn’t the first time Nvidia has backstopped Nscale
Nvidia has played guarantor for Nscale before. Around October 2025, it guaranteed approximately $860 million of Nscale’s lease obligations in Texas.
That support was tied to Nscale’s project in Ward County, Texas. The site has an IT load between 240 and 275 MW, roughly half the scale of the West Virginia building now under discussion.
Nscale has also tapped the debt markets heavily on its own. In August 2026, it closed nearly $3 billion in senior secured delayed-draw term loan facilities to fund specific GPU deployments.
AI, tech, and the markets they move—in one daily briefing.
Daily. Free. Join 34,000+ readers across crypto, finance, and policy.
Those loans were rated investment-grade. A delayed-draw facility lets a borrower pull money in stages as construction progresses, instead of taking it all upfront and paying interest on idle cash.
Then came the IPO paperwork. Nscale’s S-1 filing, which disclosed figures as of September 18, 2026, showed approximately $103.4 billion in total contracted value.
Active revenue sat around $2.6 billion. That gap between contracted value and current revenue tells the story of a company that has signed a mountain of future business and now needs to build the facilities to deliver it.
The filing also showed significant Nvidia investment participation. It outlined Nvidia’s efforts to lower Nscale’s financing costs through various guarantees tied to exclusive use of its hardware.
Nvidia’s $164.5 billion safety net
Nscale is not the only beneficiary of Nvidia’s willingness to stand behind partners. According to Nvidia’s July 2026 10-Q filing, its broader customer-support commitments carried a maximum exposure of $164.5 billion.
That reflects a deliberate strategy of helping so-called neoclouds, the newer GPU-focused cloud providers, access cheaper debt. Neoclouds want to buy enormous quantities of Nvidia hardware. Lenders hesitate to fund young companies at attractive rates. Nvidia steps in with its credit, and the chips keep flowing.
What this means
For Nscale, an Nvidia guarantee would be a meaningful advantage heading into public markets. Cheaper debt improves the economics of every megawatt it builds.
For Nvidia, guarantees tied to exclusive use of its hardware mean any facility it helps finance runs on Nvidia chips, in this case Vera Rubin.
The risk sits in that $164.5 billion exposure figure. Maximum exposure does not mean expected losses, but it does mean Nvidia’s fortunes are increasingly linked to whether its neocloud partners can keep paying their bills.
There is also the question of concentration. Nscale’s West Virginia expansion is being planned around Anthropic, so one building’s economics may hinge heavily on a single AI lab’s compute appetite.
Nvidia’s next quarterly filing should also show whether its customer-support exposure keeps climbing. For anyone tracking the AI infrastructure buildout, that number has become one of the more revealing metrics in the sector.
Disclosure: This article was edited by Diego Almada Lopez. For more information on how we create and review content, see our Editorial Policy.

