Technology

Nvidia May Back Billions in Loans for an Anthropic AI Data Center

Martin HollowayPublished 2w ago4 min readBased on 13 sources
Reading level
Nvidia May Back Billions in Loans for an Anthropic AI Data Center
source:nvidia.com

Nvidia is in talks to guarantee billions of dollars in loans for a planned AI data center in West Virginia. TechCrunch reported the talks. Bloomberg first reported on October 29, 2026 that Nvidia was discussing credit guarantees for the facility.

A data center is a large building filled with computers that run AI systems. The site is being developed by Nscale, and the loans would help pay for its electricity and construction.

The support under discussion is called a residual value guarantee. If Nscale fell behind on construction payments, Nvidia would step in and temporarily take over the facility. It would then help find a buyer and a new tenant. The talks are not final, and no terms have been agreed.

Think of it like a cosigner on a loan. A cosigner promises to pay if the main borrower cannot. In this case, the promise would come from a chip company rather than a bank.

The site is the Monarch Compute Campus, a 2,250-acre property in Mason County, West Virginia. Nscale says it will have 1 gigawatt of on-site power generation, which could grow to more than 8 gigawatts. A gigawatt is one billion watts. Nscale also calls it the first state-certified Artificial Intelligence Microgrid in the United States. A microgrid is a local power system that can run separately from the main electricity grid.

In March, Nscale, Microsoft, Nvidia and Caterpillar announced a plan to deliver 1.35 gigawatts of Nvidia Vera Rubin NVL72 GPUs to the site. GPUs are the chips that do the heavy mathematics behind AI. That same month, Nscale said it had signed a letter of intent with Microsoft for 1.35 gigawatts of AI computing capacity. A letter of intent sets out the basic terms before a full contract is written. Nscale and Caterpillar are also turning Vera Rubin reference designs, which are blueprints for building such systems, into reality on the site.

Anthropic's contract with Nscale covers the first 460-megawatt building on the campus, which is expected to use Vera Rubin chips. Nscale's IPO filing puts that contract at $44.6 billion for this one building, according to the filing with the SEC. An IPO, or initial public offering, is when a private company first sells shares to the public.

Reuters reported on October 19, 2026 that Nscale and Anthropic were in talks on a new deal worth about $45 billion. It also reported that in August, Anthropic had agreed to spend $45 billion renting AI cloud computing from the West Virginia campus. The two companies are now discussing a second deal of roughly $45 billion for another building.

Nscale is seeking to go public at a valuation of up to $35 billion. Valuation is the estimated value of the whole company. The company already runs data centers in Norway, Portugal, Texas and West Virginia. In September, Nvidia committed $1 billion in convertible financing to Nscale. That is money lent that can later be converted into company shares. That investment came before the current talks on loan support.

The broader context here is the financing structure. A chip supplier that agrees to cover construction payments and take over a facility would be carrying risks that a lender or developer would normally hold. Those are the terms being discussed, and they have not been finalized. Power and construction money largely decide when a data center starts running, so the arrangement bears directly on whether the Vera Rubin chips switch on as scheduled.

In my view, the more useful question for the industry is how many AI campuses can secure this kind of backing. Projects this size depend on a small group of large customers, a few chip makers, and a power supply that has to be delivered on time. If the Monarch campus goes ahead as described, with its own power generation and a microgrid built for the load, it would show whether projects like this can be built and financed outside the traditional utility and bank model. The outcome will depend on the final terms, the construction schedule, and how investors respond to the IPO.