Wall Street wants to pledge GPUs like homes

Nvidia and financial giant Blackstone are moving to push a new financial instrument into the market: loans backed by AI processors, structured in the same way as mortgage-backed bonds. According to Yahoo Finance, the ambition is to make graphics processing units – the expensive machines that power artificial intelligence – an accepted asset class for structured credit on Wall Street.

The idea is that data centers filled with GPUs generate predictable revenue streams from leasing computing power, and that these cash flows can be used to service bond debt. The model largely copies the logic behind mortgage securitization – with all the potential benefits and risks that entails.

"It's like issuing 30-year securities against a warehouse full of bananas" — DoubleLine Capital CEO Jeff Gundlach on AI chip securitization
Nvidia wants to turn AI chips into bank collateral – experts warn - Bilde 1

Chips age fast – that's the problem

Perhaps the most fundamental objection to the model concerns lifespan. While a home typically retains its value over decades, AI chips can become technologically obsolete within just a few years. According to research material reviewed by 24markets, experts estimate that chips are "effectively useless after 3–5 years." New generations are also "orders of magnitude more energy efficient," which further accelerates the depreciation of older models.

Meta reported a 9 percent chip failure rate in 2024 during a 54-day training run of its Llama 3 model. That illustrates that these are not passive, stable assets – they are working machines operating in a demanding environment.

$200B
Nvidia's estimated credit exposure by 2028
$60B
Broadcom's negotiated debt facility for AI chip financing
Nvidia wants to turn AI chips into bank collateral – experts warn - Bilde 2

'Phantom leverage' and hidden risk

Morgan Stanley recently initiated credit coverage of Nvidia with a neutral rating, noting that the company is increasingly using its own balance sheet to support customers' purchases of expensive hardware. The bank describes these as "new risks" that traditional accounting metrics do not capture well.

The bank's analysis suggests that Nvidia's total credit exposure – including adjustments and contingent liabilities – could amount to around $170 billion of the estimated $200 billion total by the end of 2028. Broadcom, meanwhile, is in negotiations over a $60 billion debt facility tied to AI chip financing, and is reported to have guaranteed portions of a previous $35 billion credit facility.

The liabilities are real – but they don't always show up on the balance sheet

Circular financing makes it hard to assess real demand

Another concern is what is described as circular financing: Nvidia invests in AI companies like CoreWeave, CoreWeave uses that capital to buy chips from Nvidia, and Nvidia is therefore simultaneously a supplier, a shareholder, and part of the financing ecosystem. This makes it difficult to determine whether demand for chips is market-driven or self-reinforcing.

Analysts point out that this intertwining of roles resembles the structures that helped conceal systemic risk in the lead-up to the 2008 financial crisis. Back then, it was mortgage packages with opaque underlying values – now it could be GPU packages.

The warnings from 2008 still hang in the air

According to the research material, some experts warn that a rapid conversion of bilateral credit exposures into tradeable securities, combined with vendor-controlled technological obsolescence, could amplify the correlation between defaults and collateral values – described as "a hallmark of systemic episodes."

The scenario outlined is a "2007-style run on the shadow banking system," in which doubts about credit quality trigger contractions in short-term funding and force asset sales into a falling market.

A rapidly growing market – but with unknown downside risk

That said, the outlook is not entirely bleak. The global AI hardware market was worth around $27.9 billion in 2024 and is projected to grow to over $210 billion by 2034 – an annual growth rate of more than 22 percent. Nvidia alone controls an estimated 70–80 percent of the market for discrete AI accelerators, and the company's data center revenues surged 142 percent year-over-year to $115.2 billion in fiscal year 2025.

There are also signs that older chips are holding their value longer than feared. Nvidia's A100 GPU, launched in 2020, remains in active use with contracts extending to 2029 – challenging the assumption of rapid value destruction.

The question is whether Wall Street's securitization machine can price in these nuances – or whether it will repeat the lessons of history all over again.