Nvidia announced that investors including Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR indicated willingness to commit up to $500 billion to build AI data centers. While the headline number drew attention, Nvidia’s deeper goal is to foster a secondary market for aging GPUs.
How the guarantee works
To reassure those financial backers, Nvidia has agreed to guarantee with its own funds that GPUs used as collateral in these deals will retain value. Concretely, the company promises to cover up to 25% of any shortfall if the collateralized GPUs fail to fetch their expected resale price.
Practically, if a data-center owner defaults and a lender must liquidate the chips but cannot obtain the book value, Nvidia will contribute toward the difference up to the stated limit.
Market and financial risks
The arrangement is widely described as unusual, clever and risky. A central concern is so-called “wrong-way” risk: Nvidia’s obligations would expand precisely when demand for GPUs weakens. In such a scenario, the company’s revenues could also be squeezed.
Bond markets reacted nervously to the announcement; Nvidia CEO Jensen Huang took to X and business television to clarify how the company’s exposure would be limited.
Why this isn’t simply a Lucent repeat
Some commentators compare the move to Lucent Technologies, the telecommunications-equipment supplier whose fortunes reversed during the dotcom crash after it financed customers’ purchases. Jensen Huang is aware of the comparison and says the new initiative was designed to address that concern.
The key distinction Nvidia emphasizes is that it is not bearing the bulk of capital and risk alone: the structure brings in independent, long-term institutional capital so other parties take on most of the investment and risk.
Financing context and related activity
Bloomberg has calculated that Nvidia has discussed roughly $750 billion of similar circular deals this summer. Nvidia has also previously committed billions to support customers buying its chips — including frontier AI labs such as OpenAI and Anthropic, cloud providers and specialized firms like CoreWeave (an originator of using Nvidia chips as collateral), Nebius, Firmus and Lambda.
At the same time, some traditional financing channels have become strained: several hyperscalers have taken on significant debt (for example, Oracle), issued new equity tranches (for example, Google), or burned large amounts of cash (for example, Meta). The situation has become uncertain enough that Microsoft CEO Satya Nadella recently recommended the book “1873” on a recent earnings call — a work about railroad-era financial engineering and its collapse.
Nvidia’s vision: used hardware ecosystem and long-term infrastructure
Huang argues that AI infrastructure should be viewed as investable, long-lived infrastructure — akin to railroads or airlines — rather than quickly depreciating consumer electronics. Nvidia describes its AI servers as “AI factories” that can be repurposed for another customer, cloud or operator when needs change, creating a broad potential market of users and offtakers that helps protect residual value.
If the model succeeds, Nvidia could unlock new sources of capital for AI data-center builds while sustaining demand for its hardware even as architectures age. That could expand options for startups, enterprises and researchers to access a wider array of hardware tuned to different AI needs.
Conclusion
Nvidia’s announcement creates both fresh financing avenues and novel risks. By offering a partial guarantee on future chip resale values, Nvidia aims to seed a secondary market for used GPUs, but that same guarantee exposes the company to meaningful financial pressure if AI demand falters. The outcome will depend on whether long-term, durable demand for AI infrastructure materializes and on the effectiveness of bringing in institutional capital to share the investment risk.
