NVIDIA is discussing credit guarantees for an Nscale data-center project in West Virginia that Anthropic plans to use, Bloomberg reported on October 29, 2026. The proposed support would backstop electricity and construction costs at a campus equipped with NVIDIA chips. The Information previously reported that the guarantees could reach billions of dollars, according to Bloomberg’s account.
The discussions remain private and unfinished, rely on unnamed sources, and have produced no publicly confirmed agreement. Nscale declined to comment to Bloomberg, while NVIDIA did not immediately respond. The published reporting supplies no final guarantee amount, coverage cap, maturity, or payout conditions.
NVIDIA already owns a stake in Nscale. A guarantee would deepen that relationship: the chipmaker could help make a customer’s construction financing possible, supporting a project intended to buy and operate its hardware.
That connection is the consequential part of the story. A vendor can book chip sales while accepting financial exposure that lasts beyond delivery. Understanding that exposure requires separating the value of a compute contract, the amount borrowed to build a facility, and the obligations a guarantor actually accepts.
What the Report Says About the Anthropic Project
Bloomberg describes the West Virginia campus as a flagship development for UK-based Nscale, one of the “neocloud” providers building infrastructure focused on AI computing. These companies supply access to GPU capacity, placing them between chip vendors and customers that train or run AI models.
The report also says Nscale has discussed a second $45 billion agreement with Anthropic for the campus’s second building. That is a separately discussed customer deal, not the amount NVIDIA is reportedly considering guaranteeing. Bloomberg’s account does not provide the contract duration or payment schedule needed to translate that headline value into annual revenue or near-term cash receipts.
Those distinctions matter because a large customer commitment cannot automatically pay today’s construction bills. A developer may need financing for land, buildings, electrical equipment, cooling, and other infrastructure before it can begin delivering computing services and collecting the associated revenue.
NVIDIA’s proposed support would address part of that funding problem. According to Bloomberg, the discussions cover electricity and construction costs, extending the potential backstop beyond the chips themselves.
My view is that the most useful question is which obligations NVIDIA would cover, and when lenders could demand payment. A multibillion-dollar headline says relatively little without those terms. A narrowly capped obligation tied to specific assets creates a different exposure from a broad promise covering a project’s unpaid debts.
How a Residual-Value Guarantee Would Work
A residual-value guarantee is one possible way to support financing backed by computing hardware. It should be understood as a financing mechanism, not as a confirmed description of the Nscale agreement under discussion.
“Residual value” means an asset’s expected remaining value at a future date. For a GPU system, that could mean what a lender expects to recover by selling or redeploying the equipment after several years.
A simplified arrangement would work like this:
- A lender finances equipment. The borrower uses the money to acquire systems, with those assets supporting the loan.
- The parties agree on a future value floor. A guarantor promises specified protection if the equipment’s qualifying value falls below that floor.
- The lender assesses the guarantor as well as the hardware. Stronger backing can make financing easier to obtain or improve its terms.
- A contractual trigger determines any payment. Coverage depends on the agreement’s valuation method, timing, exclusions, and maximum liability.
For an illustration, assume equipment has a guaranteed value floor of $700 million at an agreed date. If its qualifying value then falls to $500 million, the shortfall is $200 million. A guarantee might cover that difference, subject to its terms. These figures are hypothetical, not reported Nscale terms.
A residual-value guarantee differs from a general credit guarantee. The former protects a specified asset value; the latter may cover defined payment obligations when a borrower fails to pay. Neither should automatically be read as insurance against every construction delay, customer default, or operating loss.
The distinction is especially important for AI hardware. Reuters reported on October 1 that investors were skeptical about using chips as long-term collateral, and that deals in the pipeline could include stronger guarantees. NVIDIA described the assets as “productive, durable and fungible.”
The financing question is what those assets will be worth under unfavorable conditions. Hardware can remain operational while losing resale value, particularly if newer equipment offers more attractive economics.
NVIDIA’s Existing Commitments Put the Talks in Context
NVIDIA’s financial role already extends well beyond selling processors. IFR reported in its August 29–September 4 issue that the company had disclosed more than $530 billion in commitments and credit guarantees in July.
That total combines different obligations. Treating the entire sum as money lent to customers, or as expected losses, would misrepresent the exposure.
| Reported amount | What it covers | Important distinction |
|---|---|---|
| More than $530 billion | Total commitments and credit guarantees reported by IFR | Includes supplier agreements, leases, and other spending commitments |
| $108.5 billion | Credit guarantees, according to IFR | A component of the broader total |
| $105 billion | Guarantees associated with an SB Energy data center for OpenAI, according to IFR | Included within the reported credit-guarantee figure |
| $36 billion | Off-balance-sheet AI cloud agreements or backstops for neoclouds, according to CNBC | A separately described category that should not simply be added to the other figures |
IFR said $366 billion of the broader commitments related to supplier agreements, including support for factories, as well as NVIDIA’s data-center leases and research and development spending. Another $56 billion covered securing land, power, and data-center capacity for customers committed to NVIDIA chips.
CNBC’s October 24 report separately described the $36 billion in cloud agreements and said NVIDIA had signed for $20 billion in future data-center leases that would later be assigned to third parties.
These categories involve different payment triggers, counterparties, and timelines. Their overlap also makes a single grand total a poor tool for estimating downside risk. The useful comparison is between obligations with similar terms, especially guarantees that could become payable during the same downturn.
Why the Financing Is Circular
NVIDIA’s incentive is straightforward: a data center that cannot obtain funding cannot install its systems. Supporting financing can turn prospective demand into an operating facility and a hardware sale.
Frequently Asked Questions
3 questions
1Has NVIDIA Signed an Nscale Guarantee?
No signed agreement has been publicly confirmed. Bloomberg reported private, unfinished discussions over support for electricity and construction costs.
