Residual Values on Nvidia’s own graphics processing units are now being guaranteed by the company itself, in a financing consortium that has raised more than $500 billion, partnering with six Wall Street firms including KKR, Goldman Sachs, and Blackstone to turn AI compute into a bankable asset class. The arrangement effectively makes Nvidia a guarantor of its hardware’s future worth, insuring lenders against the risk that GPUs depreciate faster than loans are repaid.
The move reshapes how AI infrastructure gets built and who bears the risk of a potential AI buildout slowdown.
Key Takeaways
Nvidia has raised more than $500 billion through a financing consortium with six Wall Street firms including KKR, Goldman Sachs, and Blackstone
Nvidia reportedly guarantees lenders can recover up to roughly 25% of GPU cost if borrowers default or hardware depreciates sharply
The consortium treats entire GPU clusters, not individual chips, as the underlying asset for lending and security interests
Jensen Huang publicly described GPUs as an “investable asset class,” signaling the residual values program is a long-term strategy
Nvidia GPU Financing Turns GPUs Into Collateral
Nvidia GPU financing of this scale is new territory for the semiconductor industry. Under the structure, financial institutions extend credit to Nvidia customers, mostly hyperscalers and AI cloud providers, to purchase large quantities of H100 and Blackwell-generation GPUs.
Nvidia reportedly backstops a portion of the Residual Values of those chips, meaning it guarantees lenders can recover up to roughly 25% of GPU cost if borrowers default or the hardware depreciates sharply.
A report from August 13 put the residual guarantee at that level.
The six financial partners are acting as originators and syndicators, packaging GPU-backed loans into structured products that can be sold to institutional investors. Compute, previously treated as a depreciating operational expense on corporate balance sheets, is now being underwritten as a capital asset with an insurable floor value.
How GPU-Backed Lending Actually Works
A GPU is not like a mortgage-backed property.
It has no title, sits inside a rented rack, and its value depends entirely on software demand for its specific instruction set. Making it lendable collateral required solving a classification problem first.
The consortium’s solution is to treat the GPU fleet, not individual chips, as the underlying asset.
Lenders take a security interest in entire clusters rather than discrete units. Nvidia’s residual guarantee then functions like a put option: if the cluster is liquidated, Nvidia agrees to absorb losses beyond a set floor.
This gives lenders a defined downside, which is what structured finance requires before capital can be deployed at scale.
The mechanism is structurally similar to how aircraft manufacturers like Boeing and Airbus historically supported airline financing by guaranteeing Residual Values on leased jets. Airlines could borrow cheaply against planes because the manufacturer stood behind the resale market.
Nvidia is transplanting that model to compute, and in doing so, it is inserting itself into the capital stack of every major AI buildout it enables.
Residual Values And The Shift From Hardware Vendor To Risk Underwriter
The strategic shift matters beyond the $500 billion headline. Nvidia has historically been a capital-light business: it designs chips, contracts manufacturing to TSMC, and collects margin at the point of sale.
Taking on Residual Values exposure changes that profile in a material way.
If GPU prices fall sharply, because a competing architecture emerges, because software efficiency gains reduce the chips required per training run, or because demand simply softens, Nvidia’s guarantee could become a material liability. The company is betting that its own hardware retains value, which is simultaneously a vote of confidence in its roadmap and a concentrated directional risk.
For customers, the arrangement lowers the upfront cost of building AI infrastructure and extends the addressable market for Nvidia hardware beyond firms with sufficient capital to pay outright.
Smaller AI cloud providers and research institutions that previously could not afford Blackwell clusters may now access them through debt financing. That expands Nvidia’s customer base while deepening each customer’s financial dependence on Nvidia remaining the dominant GPU vendor.
The $500 Billion AI Infrastructure Buildout
The broader AI compute buildout provides the context.
Multiple reports from August 13 place cumulative financing commitments in the $500 billion range, with Goldman, KKR, and Blackstone among the lead arrangers. A report in FinTech Magazine noted that Jensen Huang, Nvidia’s CEO, publicly described GPUs as an “investable asset class,” signaling that the Residual Values program is a deliberate long-term strategy rather than a one-off transaction.
That framing is significant.
An investable asset class has a defined set of properties: liquidity, a price discovery mechanism, standardized documentation, and a clearinghouse for risk. Nvidia and its banking partners are building all four from scratch.
The structured products that emerge from GPU-backed lending will eventually trade in secondary markets, which means institutional investors, pension funds, and insurance companies could end up with indirect exposure to AI compute economics.
How Nvidia’s Bet Could Go Wrong
The risk is not subtle. GPU prices are not stable.
The H100, Nvidia’s dominant data-center chip through 2024 and 2025, saw spot prices fall significantly as supply caught up with initial demand. If Blackwell-generation chips follow the same curve, and if Nvidia is guaranteeing 25% residuals on a $500 billion book, the theoretical exposure runs into the tens of billions of dollars.
The analogy to aircraft leasing carries a cautionary note.
Aircraft Residual Values programs have failed before, most visibly when manufacturers over-guaranteed values on wide-body jets that airlines could not fill, forcing costly buybacks.
The difference is that aircraft demand is tied to passenger volumes, which are relatively predictable. AI compute demand is tied to model training and inference workloads, which are subject to rapid architectural change.
A single efficiency breakthrough, equivalent to what transformer architectures did to prior approaches, could strand enormous quantities of current-generation hardware.
That makes the Residual Values guarantee Nvidia has underwritten more volatile than any comparable aircraft program.
How Nvidia Got Here
Nvidia’s path to Wall Street risk partner ran through the AI infrastructure shortage of 2023 and 2024, when H100 lead times stretched past a year and cloud providers competed fiercely for allocation. That scarcity established GPUs as scarce capital goods rather than commodity components.
Cryptocurrency mining companies have made similar pivots, converting idle GPU fleets into AI inference capacity as the compute-market opportunity became clear.
The GPU financing program is a direct extension of that dynamic. Nvidia has used the scarcity period to negotiate a structural role in how its hardware gets funded, not merely sold.
Read Next: Agent Cards Are Poised to Dominate AI Payments in 2026