Nvidia's $500 billion financing plan is meeting resistance from lenders who doubt its GPUs can serve as long-term collateral, banking sources and credit managers told Reuters — with some demanding stronger guarantees than the company originally outlined.

The disagreement centers on how long Nvidia's most advanced AI chips can generate revenue. Nvidia CEO Jensen Huang said in an August blog post that the company's GPUs have a useful life of up to a decade. Credit investors and bankers see far less.

"Banks typically underwrite GPUs over a 3-4 year depreciation schedule," said Tony Trzcinka, a senior portfolio manager at Impax Asset Management. "That is different than Nvidia which argues top-tier GPUs can earn revenue for a decade."

What do lenders want instead?

Three banking sources familiar with the matter but outside the original financing group said Nvidia may need to guarantee all of its deals, or back them with revenue streams from investment-grade technology customers to cover the debt.

For now, they told Reuters, the market is not ready to treat Nvidia's compute as an investable asset comparable to aircraft — the leasing-style model the company envisioned when it announced the initiative in August with financiers including Blackstone, Apollo and KKR.

Nvidia has said some deals could carry no more than a 25% residual value guarantee, positioning the structure as a remedy for concerns about circular financing — the practice of a company helping finance purchases of its own products, which Morningstar analysts recently flagged alongside private credit and vendor financing as risks echoing the dot-com era more than 25 years ago.

Tens of billions of dollars of loan deals now in the pipeline are likely to carry strong guarantees and contracts, one source said. A second source said some structures under exploration would potentially give lenders guarantees. The deals will give AI developers access to Nvidia's compute, secured by the company's chips and backed by customer contracts plus Nvidia's underlying guarantee.

Despite the friction, demand to finance the deals remains high, the sources said. Five of Nvidia's six financial partners declined to comment; Apollo did not respond.

How long do chips actually last?

"Nvidia would imply that the GPUs work well north of five years, and that actually has been proven to be true thus far," said Andrew Chang, a director at S&P Global Ratings. "Yet we take a conservative view of the value of those chips."

Nvidia countered with third-party studies showing major cloud companies extending server depreciation periods from three-four years up to five-six. It also cited a finding from Barkr, a firm that values AI collateral such as GPUs, that its GB300 NVL72 systems could have a useful life of 9 to 10 years.

A key obstacle: there still isn't enough historical data for lenders to underwrite long-term residual value based on GPUs, the first banking source said.

"As investors, you're going to be a lot pickier about the levels that you need to get compensated for to take incremental risk," said Loren Moran, fixed income portfolio manager at Wellington Management, which has $1.3 trillion in assets under management.

An Nvidia spokesperson said its "AI compute is a productive, durable and fungible asset that can support long-term financing. Our financing partners independently assess each opportunity, including customer commitments, expected cash flow and residual value."

What precedents show

Existing chip-backed deals succeeded largely because they leaned on investment-grade revenue, not chip collateral alone:

• CoreWeave, in which Nvidia holds a stake, closed an $8.5 billion facility earlier this year — the first investment-grade GPU-backed loan — rated A3 largely because lenders rely on Meta's contractual payments.
• Broadcom backstopped more than 80% of a $35 billion financing structure supporting AI compute capacity for Anthropic.
• Nvidia itself previously provided a residual-value guarantee for financing SB Energy's Ohio data-center project, according to S&P and Moody's.

"The precedent transactions so far would suggest that the creditor community does not subscribe to long average lives for these assets," said Brian Gelfand, co-head of global credit at TCW, which manages more than $200 billion.

Huang has said he wants to make Nvidia's compute "an investable infrastructure asset," replacing the typical model of customers buying subscription contracts from AI firms. Whether the first tranche of deals closes with strengthened guarantees — and at what cost to Nvidia — will shape how easily AI developers can tap deep new pools of capital.