Nvidia tells lenders to keep backing AI servers, pitching a $500B buffer to protect GPU value

Nvidia is moving to keep capital flowing into artificial-intelligence hardware by addressing a core financing worry, the risk that expensive GPUs lose resale value as newer models arrive. In a report published Aug. 13, the company was described as laying out a plan on the order of $500 billion, designed to attract a new cohort of financiers to fund AI buildouts while mitigating concerns about what happens when older chips roll off the latest production line.

The framing in the report focuses on the economics of hardware depreciation for data-center investors. When lenders underwrite AI deployments, they are effectively exposed to whether the underlying equipment can be sold later at acceptable prices. Nvidia’s pitch, as characterized by the report, is intended to make that equipment more financeable by reducing the chance that GPUs become “aging” assets that are difficult to monetize.

A key part of the initiative is aimed at newer financiers who may not have the same tolerance for long hardware cycles as the earliest AI backers. The report suggests Nvidia wants to persuade money managers and credit providers to continue lending through the buildout period, even as fleets mature and the market moves to next-generation systems.

The article also highlights Nvidia’s attention to specific GPU configurations, including what it calls “aging GPUs,” and references the “RBOK40.” The report does not provide additional technical context in the excerpt available for this review, so it is not possible to confirm from the information at hand what the RBOK40 refers to in Nvidia’s product lineup, how it maps to data-center deployments, or how it is treated in any financing or valuation mechanism.

For Nvidia, the stakes are straightforward. The company’s revenue opportunity depends on sustained investment in data-center capacity, and the pace of those investments is influenced by the ability of customers to finance large server purchases. If financing becomes harder, capacity growth can slow, even when demand exists for training and inference compute.

The strategy also reflects a broader pattern in AI infrastructure, where hardware is frequently financed rather than purchased outright, and where lenders increasingly want clarity on residual value. In that environment, the company that can influence the confidence of the credit market can strengthen the entire supply chain, from original buildouts to upgrades that reuse or reallocate hardware.

What Nvidia disclosed beyond the reported headline figures is not clear from the materials reviewed here. The report characterizes the plan as “risky but brilliant” and centers it on lending for AI buildouts and protecting GPU value, but it is not providing, in the accessible excerpt, the structure of the program, eligibility rules, the identity of participating lenders, the terms that would govern any residual-value support, or how the arrangement interacts with customer warranties, leasing contracts, or take-back commitments.

Investors and customers will likely look for further details that specify how Nvidia’s approach changes risk allocation in AI financing. The most important items to watch are whether the initiative is tied to particular GPU families or configurations, whether it includes valuation standards for older hardware, and how quickly it can be scaled across different data-center customers. Without those specifics, the $500B figure should be treated as a reported ambition or framework rather than a confirmed, fully specified underwriting program.