Nvidia is pursuing a $500 billion programme to underwrite AI computing purchases with loans secured by its own GPUs, but Wall Street lenders are valuing the chips over only three to four years rather than the decade the company claims they can last.

The scheme, unveiled in August with partners including Blackstone, Apollo and KKR, marks Nvidia's most ambitious attempt to reshape how data-centre compute gets financed. Chief executive Jensen Huang wants computing capacity treated as a durable, productive and interchangeable asset that can underpin long-term lending. Financial institutions have not fully embraced that framing.

What is at stake for lenders?

A lender must estimate what an asset would fetch after a borrower defaults. Nvidia argues that its top-tier GPUs, including the GB300 NVL72 rack-scale systems, can keep earning revenue for up to ten years. The company points to outside research showing that large cloud providers have lengthened server depreciation schedules from three or four years to five or six years. Valuation firm Barkr estimates the GB300 NVL72 systems could remain useful for nine to ten years.

Banks and credit investors are not matching that horizon, and are operating within a more cautious framework.

How wide is the lifespan gap?

The divergence runs from roughly four years to ten. The shorter assumption translates directly into tougher lending terms: higher interest rates, thicker financial buffers and stricter repayment safeguards. Tony Trzcinka of Impax Asset Management described lenders as "considerably more cautious than Nvidia." S&P Global Ratings director Andrew Chang acknowledged that chips have so far performed well beyond five years but said his firm still values them conservatively.

A primary concern is resale value. New, faster GPU generations arrive on a near-annual cadence, and predicting the worth of any older model several years out remains difficult. The Blackwell generation that includes the GB300 NVL72 itself could be eclipsed by successors before today's loans mature.

What protections are lenders demanding?

Nvidia first suggested that some deals might carry residual-value guarantees capped at 25 percent. Bankers and asset managers now expect stronger guarantees, or backing from customer contracts and Nvidia itself, to give lenders more certainty of repayment.

The market's two largest precedents show how far those protections extend. CoreWeave, in which Nvidia holds a stake, raised an $8.5 billion investment-grade, GPU-backed loan with lenders relying largely on contracted payments from Meta. Broadcom, meanwhile, backstopped more than 80 percent of a $35 billion structure tied to computing capacity for Anthropic. Both structures lean on contracted hyperscaler revenue rather than the residual value of silicon alone.

What does this mean for AI infrastructure financing?

The arrangement is a test of whether AI infrastructure can be financed the way power plants, aircraft or ships have been for decades: as long-lived productive assets. If lenders accept longer useful lives, financing costs for AI builders could fall and the addressable market for chip-backed credit could expand toward Nvidia's $500 billion target. If lenders stay anchored to shorter horizons, the burden falls back onto balance-sheet guarantees and customer contracts.

The outcome hinges on whether real-world evidence of chip longevity can close the gap between Nvidia's decade-long outlook and lenders' shorter horizons. Until then, borrowers are likely to absorb the cost through stronger guarantees and richer pricing.