The disagreement comes down to a simple question: how long does a graphics processing unit keep earning money? Nvidia says its top processors can generate returns for a decade. Banks and credit investors, the people who would lend against them, are working from a shorter clock of three to four years. The gap between those two answers is holding up a financing plan built around as much as $500 billion of chips.
Nvidia is trying to raise the money using its own GPUs as collateral, and on October 1 the plan ran into hesitation from lenders, Reuters reported, citing people involved in the talks. Banks and credit investors are asking for thicker protection than Nvidia has offered so far, the people said. The negotiation is not about whether Nvidia can borrow. It is about what happens if the collateral loses value.
At the center of the dispute is a disagreement about depreciation. Nvidia argues that its top processors can keep making money for a decade, because the demand for computing has kept rising and because the chips can be rented, resold, or repurposed as workloads change. Banks typically assume a three-to-four-year useful life, a conservative window rooted in how quickly successive generations of chips have made the last ones obsolete. Lenders price their risk from the short clock; Nvidia wants credit for the long one.
The stakes are concrete. For multi-billion-dollar deals, lenders are expected to demand supplemental customer contracts or repayment guarantees before they will sign, according to the report. That is a way of tying the loan’s value to something steadier than a chip’s resale price. A lender asked to accept GPUs as collateral wants to know that, if the borrower cannot pay, there is a stream of revenue somewhere behind the hardware.
The thicker the protection Nvidia is asked to provide, the more constrained it becomes. Reuters noted that the more collateral the company pledges, the less room it has to maneuver if chip prices fall or returns on AI infrastructure weaken. A company that has borrowed against its own inventory has fewer assets left to use as a cushion, and it owes money precisely when the market that secures the loan is turning down.
The financing plan is a measure of how central Nvidia has become to the credit markets, not just the computing markets. GPUs, once a niche purchase, now underpin hundreds of billions of dollars of value across the industry. When a company can propose using its own chips to raise $500 billion, the hardware has stopped being merely a product and has become a financial instrument.
That transformation cuts both ways. On the way up, chips as collateral make borrowing cheap and expansion fast. On the way down, the same chips become a liability, because their value is tied to the very demand that a downturn would undermine. Lenders understand this cycle better than most, which is why they are asking for more than Nvidia has offered. They have been through enough turns to know that a ten-year projection looks different from a three-year one.
Nvidia’s position reflects its confidence in the durability of the AI build-out. The company has argued that the demand for its processors is structural, not cyclical, and that a chip bought today will still be earning money years from now. The banks are not disputing the demand. They are disputing the resale value, and in a collateral negotiation, resale value is what counts.
The talks will come down to terms. Lenders want customer contracts and repayment guarantees attached to the largest deals, and they want depreciation schedules that reflect their view of how fast the hardware loses value. Nvidia wants recognition that its best chips hold value longer than the standard schedule allows. Somewhere between the two is a deal, and the size of the concessions on either side will determine how much of the $500 billion is actually available.
Nvidia’s willingness to borrow against its own chips is a sign of how much capital the AI build-out still demands. The company has been at the center of a spending wave that has stretched balance sheets across the industry, and even the maker of the most sought-after hardware has found reasons to finance rather than pay as it goes. The question the lenders are asking is the one every creditor eventually asks: if the cycle turns, will the asset still be worth what was lent against it. The answer, they have concluded, depends on terms Nvidia has not yet agreed to.
The outcome will be read across the industry. Other companies that bought GPUs in bulk are watching, because they too may want to borrow against their hardware. If lenders succeed in tightening the terms, the cost of that borrowing will rise. If Nvidia holds the line, the market will have set a precedent for treating chips as long-lived collateral. Either way, the question of how long a GPU keeps earning money is now a matter for bankers as much as for engineers.
