Wall Street is telling Nvidia that its chips are valuable — just not valuable enough, for long enough, to underwrite a half-trillion-dollar financing vision on the terms the company sketched in August.
In a Reuters Focus report dated Oct. 1, 2026, banking sources and credit managers said lenders doubt Nvidia graphics processors can serve as long-term collateral the way the company describes. They want richer guarantees on chip-backed loan structures tied to Nvidia’s plan to help mobilize more than $500 billion of third-party capital for AI infrastructure. Pipeline deals, sources said, are already being shaped with more certainty for investors — including stronger guarantees and customer contracts — even as demand to finance those deals remains high.
The friction matters because the AI buildout is increasingly a capital-markets problem. Hyperscalers, neoclouds and model labs need accelerators, power and shells at a scale that stretches balance sheets. Nvidia’s August answer was to treat “compute” as investable infrastructure — closer to aircraft leasing than a one-off equipment sale — and to bring in six of the largest names in alternative capital and banking.
On Aug. 10, Nvidia said it had signed memorandums of understanding with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR to establish independent compute financing platforms designed to mobilize **over $500 billion** of third-party capital **over time**. The partnerships remain subject to final agreements.
A day later, CEO Jensen Huang clarified the figure on Nvidia’s blog: that sum is aggregate capital those platforms are designed to mobilize. It is **not** Nvidia revenue, **not** a single fund and **not** a commitment to one customer. Partners would independently underwrite each opportunity. To answer “circular financing” critics, Huang said Nvidia might, in some cases, provide residual-value support for **up to 25%** of an opportunity, assessed project by project — limited support meant to complement independent underwriting.
That package — a big headline number, MoUs rather than closed funds, and modest optional residual support — is what credit markets are stress-testing.
On-record sources frame a depreciation and recovery gap. Tony Trzcinka, a senior portfolio manager at Impax Asset Management, told Reuters Wall Street is “much more conservative” than Nvidia’s claim that top-tier GPUs can earn revenue for a decade. Banks, he said, typically underwrite GPUs on a **three- to four-year** depreciation schedule; investors will likely demand higher rates, thicker cushions and stronger repayment protections.
Andrew Chang of S&P Global Ratings said Nvidia’s implication that GPUs work well beyond five years “has been proven to be true thus far,” while still taking “a conservative view of the value of those chips.” Brian Gelfand, co-head of global credit at TCW, said precedent deals suggest creditors “do not subscribe to long average lives for these assets.”
Three banking sources outside the original financing group told Reuters the market is not yet ready to treat Nvidia compute like aircraft. Nvidia may need guarantees on all deals, they said, or structures backed by investment-grade customer revenue. Five of Nvidia’s six August partners declined to comment; Apollo did not respond.
Nvidia disputed the durability critique without answering the expanded-guarantee question directly. A spokesperson said AI compute is “a productive, durable and fungible asset that can support long-term financing,” with structures that “will vary as this market develops.” The company pointed to longer cloud server depreciation periods and a Barkr finding that recent GB300 NVL72 systems could have a nine- to ten-year useful life.
Recent GPU-backed loans show why cash-flow certainty still dominates collateral. CoreWeave’s roughly $8.5 billion facility was rated A3 largely because lenders could rely on Meta’s contractual payments, Reuters reported. Broadcom backstopped more than 80% of a $35 billion Anthropic-related structure — far heavier support than Nvidia’s optional 25% residual framing.
Chips can secure a deal. Lenders still want to know who pays if utilization or resale value disappoints.
There is no public evidence that $500 billion has been committed, and no standardized platform term sheet. It is not established that any MoU partner has walked away. Anonymous sources say tens of billions of pipeline loans are likely to carry stronger guarantees; Nvidia has not confirmed expanded support. A separate residual-value guaranty tied to Ohio data-center arrangements should not be conflated with the August financing-platform ambition.
For now, the story is less “Wall Street kills the plan” than “credit markets renegotiate it.” Nvidia wants GPUs treated as long-duration infrastructure. Lenders are still underwriting them closer to depreciating tech equipment — and asking who covers the residual if that story is wrong. That pricing fight is itself an AI-capex bottleneck.