What a GPU will be worth at the end of a financing term shapes how that financing is structured.
Every GPU financing rests on a view of what the hardware will be worth when the term ends, and getting that view right separates workable terms from unworkable ones.
Residual value is the estimated worth of an asset at the end of a defined period. In most equipment financing, the estimate is routine. For AI hardware, the estimate is one of the most debated numbers in the deal.
GPUs can lose value in ways that are hard to forecast. New generations arrive quickly, rental prices move with supply and demand and the secondary market is still maturing. How residual value is modeled, and whether a real path exists to realize it, affects term length, structure and whether a deployment can be financed at all.
Residual Value Determines How Much Risk a Financing Carries
In any asset-backed financing, the provider is exposed to the value of the asset if the borrower cannot repay. The higher and more predictable the residual value, the more comfortable a financing provider can be with longer terms and more flexible structures. When residual value is uncertain, providers protect themselves by shortening terms, requiring larger upfront payments or declining.
For GPUs, residual value is a central issue because each new architecture can reduce the value of existing hardware, and estimates of how much vary widely. A financing structure that ignores that variation will either misprice risk or be too conservative to be useful.
Hardware Generations Drive Much of the Change in GPU Value
Each new generation delivers more performance per dollar and per watt, and that gain reduces what buyers will pay for the previous generation. Pricing data also shows how well prior-generation hardware holds up: at its launch, Hashrate Index’s AI Hardware Price Index put refurbished eight-GPU H100 nodes at a median asking price of about $246,000, or 76 percent of new H100 node asking prices, against about $544,000 for a new B300 node.
Rental prices move in both directions. One-year H100 contract pricing rose almost 40 percent between October 2025 and March 2026 as on-demand capacity sold out, according to SemiAnalysis, and the median on-demand H100 price was about $3.25 per hour in September 2026, according to CloudZero. Rental income supports resale value when it holds up and weakens it when it softens.
Useful Life Is Debated Among Cloud Providers and Critics
Older GPUs often stay in productive use for years. Google Cloud ends support for NVIDIA P100 GPUs on September 15, 2026, several generations after the chip launched, according to Google Cloud documentation. Alphabet depreciates servers over six years, Meta over five to five and a half and Microsoft over a range of two to six, according to GPU Insights’ review of annual reports.
Amazon shortened the schedule for a subset of its servers from six years to five effective January 2025, and some critics argue the real economic life is closer to two or three years, according to CNBC. For financing, useful life should be modeled with care and tested against remarketing data.
How Residual Value Is Modeled
Residual value modeling estimates what a specific piece of hardware is likely to be worth at each point in its expected operational life. A sound model considers the hardware generation and its position in the product cycle, the expected utilization and workload type, physical condition and maintenance history, observed secondary market and rental pricing and the expected timing of the next generation.
Public price data has limits: listed prices are snapshots of what sellers ask, private transactions can clear at different levels and data sets disagree. CloudZero lists secondary-market H100 cards at $25,000 to $35,000, for example, while Servnet reports used H100 cards at roughly $12,000 to $22,000. Models are most reliable when they draw on actual remarketing activity in addition to published asking prices.
Secondary Market Placement Turns Residual Value Into an Exit
A residual value estimate reduces risk only if the hardware can be sold at the end of its term. Secondary market placement, the ability to remarket and resell retired GPUs to other buyers, gives the estimate an exit.
A secondary market already exists for prior-generation hardware, and Hashrate Index tracks refurbished H100 nodes alongside new ones. A financing provider with access to that market can assign value to retired hardware that a conventional equipment schedule might write down quickly.
How Residual Value Shapes the Financing Structure
Residual value influences most terms in a GPU financing. A credible residual value supports longer terms aligned with the hardware’s operational life and spreads payments more evenly while the cluster ramps up. Residual value also affects whether a lease or an asset-backed loan is the better fit and how refresh and replacement are planned at the end of the term.
GPUFinancing, one of Vertical Data’s three platforms, arranges structured financing for GPU deployments. Vertical Data’s managed infrastructure services then support the hardware across its operational life.

