AI Circular Financing Grows Larger: The Key to Success Lies in NVIDIA GPU 'Residual Value'

Wall Street is increasingly investing in AI infrastructure, but the sustainability of this large-scale 'circular financing' model rests on a key assumption: that NVIDIA (NVDA-US) GPUs will not depreciate as quickly as past technology equipment.

As NVIDIA partners with Wall Street financial giants to launch a $500 billion financing framework, the lifespan and residual value of GPUs are shifting from a technical issue to the core determinant of whether the entire AI financing system can hold.

Under this framework, technology companies can lease NVIDIA chip equipment with financial backing from major institutions such as Apollo Global, KKR, Brookfield Asset Management, BlackRock, Blackstone, and Goldman Sachs.

Senior financial executives involved in the deals believe that, given the ongoing supply shortage of AI chips, chip prices may remain high longer than many analysts initially expected.

NVIDIA CEO Jensen Huang described the arrangement as creating a new asset class based on chips, opening investment opportunities for the $22 trillion private capital industry.

GPU Longevity Becomes the Financing Key

The problem is that GPUs are not real estate, nor are they like commercial aircraft or automobiles, which have relatively stable secondary markets even if lessees default. With rapid AI technological iteration, the economic value of older chips could plummet quickly if demand cools, model efficiency improves, or a new generation of GPUs makes a significant performance leap.

Ben Bajarin, a technology analyst at Creative Strategies, stated bluntly that the entire model relies on continuous investment; risks will emerge if overcapacity occurs or computing demand weakens.

Currently, many debts secured by GPU leases require full repayment within 3 to 5 years, based on the assumption that the chips' residual value may be negligible after that period.

However, companies purchasing or leasing GPUs must also invest heavily in building data centers, where revenue recovery may lag far behind initial expenditures, making asset depreciation a significant risk in the financing structure.

The model's concept involves issuing public and private bonds to institutions such as pension funds, insurance companies, and sovereign wealth funds, then channeling the funds into specially established platforms to continuously finance AI chip transactions.

The issue is that if the AI infrastructure construction boom stalls, projects are delayed or canceled, or NVIDIA itself launches a breakthrough product, the residual value of older GPUs could decline faster than financial institutions anticipate.

Older A100 Chips as a Residual Value Benchmark

Despite the risks, financial institutions currently involved in the transactions remain relatively optimistic. They believe that NVIDIA GPUs are currently in severe supply shortage, giving them better collateral characteristics than traditional computer processors.

Huang also emphasized that many customers continue to use early-generation chips, proving that older GPUs do not rapidly lose value upon the release of new models.

A recent case from CoreWeave (CRWV-US) serves as strong evidence. The company, which primarily purchases NVIDIA GPUs, builds data centers, and rents out computing power, recently signed an A100 chip lease contract at nearly full price, with the contract extending all the way to 2029.

Notably, the A100 was launched as early as 2020, yet it still commands lease prices close to those of new products, suggesting that the economic lifespan of some older-generation AI chips may be longer than originally estimated by the market.

Brannin McBee, Co-Founder and Chief Development Officer of CoreWeave, stated that rising demand for inference and engineering workloads is extending the useful life of older GPUs.

As AI computing expands from model training to inference and AI agent applications, even non-latest-generation chips may find sufficient demand.

NVIDIA Personally Backstops GPU Residual Value

To further reduce financial institutions' concerns about depreciation, Huang revealed that NVIDIA may, in certain cases, provide a 'residual value support mechanism,' using its own balance sheet to support up to 25% of project costs, with a total cap of approximately $125 billion. If the value of the hardware used as collateral falls below a preset level, NVIDIA and other guarantors must contribute funds to cover the shortfall.

Senior executives involved in the $500 billion financing deal noted that NVIDIA has up to 10 years of historical unit lease price data for GPUs, giving it the ability to assess asset values. The residual value protection offered by NVIDIA can be seen as the 'first loss layer' in the entire financing structure, absorbing unexpected hardware depreciation risks first.

With this layer of protection, private capital firms gain greater confidence in packaging GPU lease financing into standardized securities and selling them to long-term debt investors such as insurance companies, further unlocking the vast capital pools of institutions like Apollo, KKR, and Brookfield.

For NVIDIA, this structure is equally attractive, as it allows financing risks and financial burdens to be shifted more to financial consortia rather than being entirely borne on its own balance sheet.

In other words, the continued expansion of AI 'circular financing' increasingly depends on a seemingly simple but difficult-to-answer question: How much will today's multi-thousand-dollar GPUs be worth in 3, 5, or even more years?

As long as older chips can maintain leasing demand and residual value, Wall Street will have the opportunity to transform GPUs into financeable, securitizable assets.