At the center of the model is a question that matters to both Nvidia and its financing partners: how quickly does an AI chip lose its economic value?
Huang argues that Nvidia’s systems should be viewed differently from conventional technology equipment. “Nvidia’s AI factory platform is really an investable asset, an infrastructure asset,” Huang said. “The reason for that is because it’s productive, it’s revenue generating, it is fungible, it’s used by just about every cloud service provider, it runs every AI model.”
Recent rental pricing supports part of that case. One-year rental rates for Nvidia’s H100 increased from roughly $1.70 per GPU-hour in October 2025 to $2.35 in March 2026. Median on-demand pricing across providers moved from about $2.00 per GPU-hour in October 2025 to $2.70 in June 2026. Newer B200 capacity carries higher rates of roughly $5.30 to $7.05 per GPU-hour.
Nvidia also points to the longevity of older hardware. Its A100, introduced in 2020, remains commercially active across training, inference, fine-tuning and high-performance computing workloads. “Customers continue to commit capacity for multi-year deployments, extending A100's economic life toward a decade,” Huang said.
Software is another part of Nvidia’s residual-value case. The company argues that improvements to CUDA can increase the productivity of GPUs already in service, helping installed hardware remain useful even as newer generations become available. Because Nvidia’s architecture is deployed across major cloud providers and supports a broad range of AI workloads, Huang also contends that compute capacity can move among customers and applications as demand changes. “In AI, compute is revenue,” Huang wrote.
That durability is particularly important for asset-backed financing. Investors lending against physical infrastructure typically have collateral they can recover and sell if a borrower defaults. Financing GPUs creates a less established version of that model because the future resale value and useful life of advanced chips remain uncertain.
Ben Emons, founder of FedWatch Advisors, identified depreciation as the central vulnerability. “Depreciation is the one key risk here,” Emons said, warning that Nvidia chips “could depreciate faster than expected.” Older GPUs can shift from demanding model-training workloads toward lower-margin inference as newer hardware arrives. That transition can affect both the income a GPU produces and the price investors could recover by selling it, making assumptions about residual value important to the financing structure.
Nvidia plans to provide residual-value support covering as much as 25% of an opportunity when warranted, while the participating financial firms retain responsibility for underwriting individual projects. The prospective customers include AI labs, startups, enterprises, cloud providers and sovereign buyers.
Borrower quality adds another layer of risk. A Bank of America Securities note said prospective borrowers are likely to include non-investment-grade companies, including AI startups and neocloud providers that have limited access to conventional debt markets. A default could therefore leave investors attempting to sell repossessed GPUs while the value of that equipment is declining.
Emons estimates investors could respond by treating GPUs more like rapidly depreciating equipment than long-lived real estate. Under that scenario, he expects required returns could fall in the 11% to 17% range depending on an investor’s position in the capital structure.
China represents another potential challenge to those assumptions. Emons said greater Chinese production could eventually create pricing pressure if manufacturers put large volumes of lower-cost chips into the market. A significant decline in hardware prices could reduce the collateral value supporting GPU-backed loans faster than the associated debt is repaid.
There are currently significant constraints on that scenario. Huawei, a major Chinese AI-chip supplier, has been on the U.S. Commerce Department’s Entity List since 2019. The U.S. government also said in May that Huawei Ascend AI chips violate U.S. export controls, preventing American companies from using them. Nvidia, meanwhile, remains the dominant U.S. supplier of AI chips, with market share estimated above 75%. Scarcity and continued demand for computing capacity have also supported rental economics for its existing hardware rather than producing the rapid decline that would undermine the infrastructure thesis.
The $500 billion target represents capital that the six financing platforms are designed to mobilize over time. It is not Nvidia revenue, a single investment fund or funding committed to one customer. Each participating firm will make its own decisions about which projects justify financing. That distinction makes the initiative as much a test of Nvidia’s residual-value argument as a mechanism for expanding AI infrastructure. Huang is asking institutional investors to underwrite GPU systems on the premise that their productive life, software ecosystem and broad customer base can support years of cash generation.
If those assumptions hold, institutional financing could make large AI deployments accessible to customers that cannot fund them entirely themselves. If depreciation accelerates, utilization weakens or competing hardware sharply reduces prices, investors could instead discover that the collateral behaves much more like technology equipment than traditional infrastructure.
The scale of Nvidia’s financing push means the answer will matter beyond the useful life of any individual GPU. Hundreds of billions of dollars in potential investment now depend in part on whether AI compute can retain enough economic value to support the long-duration financing Huang wants Wall Street to provide.
This analysis is based on reporting from CNBC.
Image courtesy of NVIDIA.
This article was generated with AI assistance and reviewed for accuracy and quality.
