Nvidia’s new plan is not limited to financing up to $500 billion to build AI data centers; its deeper purpose is an attempt to create a secondary market for graphics processing units as they age, so that these chips do not rapidly lose their value and remain capable of attracting financing and use.
Nvidia announced this week that Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR are prepared to commit up to $500 billion to build AI data centers. But to persuade these major financial institutions, the company agreed to guarantee the value of its chips used as collateral in these transactions, using its own funds.
A Guarantee for the Value of the Chips
Under the plan, Nvidia will cover up to 25% of the shortfall if the graphics processing units used as collateral do not retain their expected value. If a data center owner is unable to repay a loan and the lender is forced to liquidate the assets, and the chips cannot achieve the price recorded on the books, Nvidia will pay part of the loss within this limit.
Julie Bort, editor of TechCrunch’s startups and venture capital section, sees this move as unusual, smart and dangerous at the same time. The plan has raised concern in the bond markets to the extent that Nvidia CEO Jensen Huang was prompted to explain the limits of the company’s risk through X and business-focused media outlets.
One risk point is what financiers call “wrong-way risk.” Nvidia’s obligations will increase at the same time that demand for its chips weakens, a scenario that could also put pressure on the company’s revenue.
A Bet on Reusable Infrastructure
The plan links data center financing with maintaining demand for Nvidia hardware as it ages. According to the analysis, Huang wants a broad ecosystem for used hardware to emerge, allowing startups, enterprises and researchers to access multiple types of equipment, each suited to different AI needs.
Huang describes Nvidia’s servers as “AI factories,” comparing them to railways or airlines rather than rapidly consumed assets such as personal computers. In his view, when needs change, a computing factory can move to another customer, another cloud or another operator. This broadening of the base of potential users and end buyers could help protect the residual value of computing capacity.
If this vision is realized, Nvidia will care not only about new architectures, but also about the ability of older architectures to continue operating within a broader market. Bort believes this could give startups, enterprises and researchers an opportunity to use less expensive hardware, much as they choose open-weight AI models alongside advanced models.
The Lucent Comparison and the Risks of Circular Financing
The comparison with Lucent Technologies casts a shadow over the plan. Lucent was a supplier of telecommunications equipment that rose and then collapsed with the dot-com bubble after lending to its customers to buy its products. Huang is familiar with this comparison, and wrote on X that the new initiative is designed to address this concern by bringing independent, long-term institutional capital into the AI infrastructure market.
The plan differs from Lucent’s model, according to the analysis, in that Nvidia is not directly bearing most of the capital and risk, but is instead trying to attract other institutions while pledging to protect part of the value of its chips in the future. Nevertheless, the company continues to provide billions of dollars to those buying its chips, including AI labs OpenAI and Anthropic, and new cloud companies CoreWeave, which originated the idea of using Nvidia chips as collateral, as well as Nebius, Firmus and Lambda. Bloomberg also calculated that Nvidia was working this summer on other circular deals worth $750 billion.
Will Demand for AI Continue?
The plan comes at a time when some traditional financing methods have begun to weaken. Some hyperscale computing companies have taken on more debt, such as Oracle, issued new stock tranches, such as Google, and spent large amounts of cash, such as Meta. Microsoft CEO Satya Nadella also recently recommended the book “1873” during the company’s latest earnings call; the book addresses financial engineering in the railroad era and the economic collapse that followed it.
The fundamental risk is that the current AI boom, in which demand exceeds available capacity, may not continue. Consumers and enterprises may reduce their use, or technologies may emerge that make current infrastructure more efficient or render part of it obsolete. Demand for servers and chips could then dry up, causing the value of the collateral to decline and exposing Nvidia to dual pressure on its obligations and revenue.
Bort concludes that Nvidia’s strength and the time window available to it give the company an opportunity to try to build this system. Success would mean new financing sources for constructing AI data centers and a broader market for aging hardware; failure would tie the company more closely to the ability of demand for AI to continue.