Technology

Nvidia Says It Will Only Back Data Center Loans Once the Rest of the Financing Is in Place

Martin HollowayPublished 3m ago4 min readBased on 1 source
Reading level
Nvidia Says It Will Only Back Data Center Loans Once the Rest of the Financing Is in Place
Photo by Brett Sayles on Pexels

Nvidia says a loan guarantee takes effect only after the customer has secured the financing needed to build the facility and install Nvidia's products. The company also said that guaranteeing a loan does not mean Nvidia is providing the financing itself. The clarification came on November 18, 2026, alongside third-quarter results showing net income of $64.9 billion, up from $59.7 billion in the prior quarter.

The clearest test of the policy is Nvidia's largest loan-guarantee deal, a $105 billion backstop for OpenAI's PORTS-Pike data center campus in Ohio, announced in August. In exchange for the guarantee, OpenAI promised that the campus would use Nvidia equipment exclusively. The guarantee has been reported to have been cut from an earlier figure of $250 billion to $105 billion, a change attributed to Colette Kress, Nvidia's CFO who is now interim CEO.

Kress's wording on the conditions is blunt. If a project is delayed or runs into difficulty during buildout, Nvidia's guarantee "doesn't exist" anymore. She described the clarification as a restatement of existing terms rather than a change in policy.

Kress also said Nvidia will be more selective about financing data centers. The company will prioritize the availability of land, power and shell, meaning the bare building envelope, and it will scrutinize the types of companies it helps. Those criteria point to where Nvidia's credit exposure will and will not go in the coming period, though the company has not set out a specific number of projects or a dollar ceiling.

The broader context here is the difference between a guarantee and a loan. A guarantee is a contingent obligation, a promise that pays out only if a defined event occurs. It costs Nvidia nothing unless the underlying project fails to be financed or built, and it becomes a claim on Nvidia's balance sheet only if the customer's own capital stack, the mix of debt and equity funding the project, holds together through construction. Kress's language makes the trigger explicit: the backstop is tied to project execution, not to the announcement of a campus. For anyone modeling Nvidia's exposure, the operative question is how many of its guarantees sit at each stage between signed financing and energized hardware.

For the people who build these facilities, the practical signal is different. Prioritizing land, power and shell availability moves the qualifying bar upstream, toward the constraints that most often stall large sites, such as interconnection queues for grid connections, substation capacity and permitting. Conditioning support on those factors could mean fewer projects that stall halfway through construction with GPUs ordered and no power to run them.

In my view, the exclusivity term deserves as much attention as the guarantee itself. Tying a backstop to a single vendor's equipment gives the vendor a strong position in how the campus is specified and supplied, and it constrains the customer's hardware choices for as long as the arrangement holds. Whether that trade-off serves both parties over the life of a multi-year build will depend on how the project actually unfolds.

Nvidia's stated intent is narrower than the headline numbers suggest: fund less speculatively, and stand behind projects only once the rest of the financing is in place. If that discipline holds, it gives the wider buildout a clearer sequence of who carries risk and when, which is useful to anyone trying to plan capacity against real delivery dates.