Wall Street is fighting over a number almost nobody can verify: what a used Nvidia GPU is worth when a lender needs its money back. The answer determines whether Nvidia’s $500 billion AI financing ambition is a structural breakthrough or a heavily backstopped bluff.
The Bull Case: Chips That Earn Like Infrastructure
Nvidia asked Wall Street to treat its graphics chips like real estate, unveiling financing platforms with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR aimed at mobilizing more than $500 billion in third-party capital for AI infrastructure over time. The bullish logic flows from one empirical claim: that Nvidia hardware actually lasts.
Nvidia has argued that older systems can remain in active commercial use years after launch, citing continued use of the A100 years after its 2020 introduction, and it has said that, in some cases, it may provide a residual-value support mechanism for up to 25% of an opportunity on a project-by-project basis. If those claims hold for Blackwell-generation hardware, Nvidia’s case is that accelerated computing can function more like productive infrastructure than ordinary equipment, with its systems broadly adopted, transferable among customers, and supported by CUDA software, which it argues can extend their useful economic life.
Inference workloads are a major reason older accelerators may keep earning, because they tolerate previous-generation silicon far better than large synchronized training runs do. A chip that trains today can infer tomorrow. That migration path is the backbone of the bull case.
The Bear Case: Energy Math Destroys the Model
Credit investors say banks commonly underwrite GPUs over a three-to-four-year depreciation schedule, while Nvidia argues its leading systems can keep earning money much longer. The bears are not wrong on physics. When a newer chip delivers far better performance per watt, data center operators cannot justify occupying scarce megawatts with older, power-hungry silicon. The residual value of older chips can fall not because they stop computing, but because their hosting cost exceeds their economic output.
Lower collateral values could mean more equity upfront, tighter lending terms, or higher borrowing costs. Multiply that across a $500 billion target and the arithmetic turns punishing fast. The distinction between a platform and a funded loan is important: Nvidia described the $500 billion as a mobilization target across proposed platforms over time, not debt outstanding. Nvidia’s own residual-value support, when offered, covers only up to 25% of an opportunity, and the company has described it as something it may provide in some cases on a project-by-project basis.
What the Broadcom Structure Actually Tells Us
The most instructive data point in this debate is not Nvidia’s support. It is Broadcom’s.
Broadcom established the AI XPV Platform with Apollo and Blackstone to enable more than 20 gigawatts in compute capacity through 2028, launching with an initial tranche of $35 billion led by Apollo, in partnership with Blackstone, to facilitate Anthropic’s previously announced expansion of more than one gigawatt of compute infrastructure expected to begin deploying in Fluidstack-based sites starting in mid-2026.
Broadcom has also disclosed that the structure includes a backstop tied to a customer’s lease obligations over five-year lease terms, with the total backstop amount increasing as AI racks are delivered and decreasing as the customer makes payments.
The pricing effect of that backstop is striking. Reporting on the structure has described a senior tranche priced at 5.75% that benefited from Broadcom’s backstop, alongside a junior tranche priced at 8.5% that did not. Nearly three full percentage points of borrowing cost separated backed financing from unbacked financing. That spread is the market’s verdict on standalone GPU collateral: it does not yet stand on its own.
Broadcom is still on the hook if its customers cannot pay, even with the debt structured off-balance-sheet. As CNBC reporter Kristina Partsinevelos said on air in August 2026, Broadcom is guaranteeing part of the debt, and that guarantee is where the risk sits.
Where the Evidence Leads
The Broadcom structure does not vindicate the six-year bull case. It concedes the three-year bear case is real enough to price around. Lenders demanded a strong corporate backstop before accepting rates anywhere close to investment-grade. Nvidia’s offer to provide residual-value support of up to 25% of an opportunity, project by project, may fall short of what moved the needle in the Broadcom deal.
The decisive question is not whether an Nvidia GPU still works after five or six years. It is whether that chip can earn enough revenue, at sufficient utilization, to satisfy financing obligations throughout that period. Physical longevity and economic longevity are not the same thing, and the debt market is currently pricing them as very different risks.
Watch two indicators: whether Nvidia expands its residual support beyond 25% to attract senior lenders, and whether any of the six August 10, 2026 platform partners close an actual financing transaction before year-end. Until then, the $500 billion remains a target, not a loan book.
