Nigel Green On What Nvidia’s $500bn Deal Leaves Out

The announcement

Nvidia said on August 10 that it had signed memorandums of understanding with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR to establish independent compute financing platforms. The stated aim is to mobilize more than 500 billion dollars of third-party capital for the buildout of AI infrastructure over time, and to create dedicated pools of capital for Nvidia customers at rates the company describes as attractive.

Nvidia is pitching compute as an asset class. Jensen Huang said the company had gone from building chips to creating a new category of investable infrastructure, which the release calls AI factories. “In AI, compute is revenue,” he said.

Executives at all six firms gave supporting statements, calling compute a scarce and mission-critical asset and pointing to the capital the buildout will need. The release ends by saying the partnerships remain subject to execution of final agreements. Nvidia published no terms, no individual commitments and no timeline.

Why Nvidia says the chips will hold their value

Anyone lending against GPUs over a long term needs a view on what the hardware will fetch when the term ends. Nvidia’s answer is its software.

The company says its compute is widely adopted and can be moved between customers and operators, and that steady improvement through CUDA stretches the useful life of the hardware and improves its economics over time. The release says Nvidia compute offers the longest life in its category, along with the lowest token cost and the highest revenue.

Everything rests on that. If the chips hold their value, lenders can underwrite compute over long terms.

Nigel Green’s response

Nigel Green, chief executive of deVere Group, took issue with the deal two days later. He points to a provision he says commits Nvidia to guaranteeing up to 25 percent of the residual value of its own chips in individual financing transactions.

“If demand for Nvidia’s chips is genuinely as strong and durable as the market currently believes, why does the company need to personally guarantee the resale value of its own hardware to get lenders comfortable?”

“Confident sellers don’t usually need to underwrite their own customers’ financing. Companies that do this are, typically, trying to solve a problem the headline numbers are not showing.”

The size of the platform, he says, distracts from what it took to put together.

“A $500 billion financing platform sounds like strength, and it’s being widely reported that way,” he says. “I say, look more closely at what it actually requires. Nvidia is actively engineering the conditions for that demand to be financeable at scale, routing the vast majority of it through private credit, a market with considerably less transparency than public lending.”

Green then turns to how quickly the chips are replaced.

“Nvidia introduces new GPU architectures roughly every two to three years,” he says. “A lender being asked to finance chip purchases over a much longer horizon needs real confidence that the hardware retains value well beyond that replacement cycle.”

“The fact Nvidia felt it necessary to personally guarantee a quarter of that residual value suggests even Nvidia is not fully confident the market would extend that credit on the hardware’s merits alone.”

What the announcement leaves out

That provision is not in Nvidia’s announcement. The release makes no mention of residual value, depreciation or a guarantee, and the figure of 25 percent does not appear anywhere in it. What the document contains is the six partnerships, the capital target, Huang’s case for compute as an asset class, statements from each firm and a line saying final agreements have yet to be executed.

The figure has been widely reported and traced to separate comments Huang made around the announcement, which describe the support as optional, capped per project and assessed case by case. NervNow could not check those comments against anything Nvidia has published.

Green’s statement presents the provision as confirmed alongside the memorandums of understanding.

What nobody can settle yet

Green stops short of calling the buildout a bubble.

“Of course, none of this means the AI infrastructure buildout is not real, or that Nvidia is in trouble.”

“However, it means the company underwriting its own customers’ debt, on the value of its own product, might be a sign of a market not, perhaps, brimming with confidence.”

He ends by telling anyone exposed to Nvidia, to the private credit funds involved or to the wider AI trade to go and investigate the platform.

“This does not mean history repeats exactly, but it means the underlying mechanics deserve the same level of scrutiny they would receive in any other sector.”

The announcement cannot settle the argument. Nvidia says software keeps its hardware working longer. Green says a backstop would not be needed if lenders believed that. Nvidia’s claim is in the document. The evidence for Green’s is not, and the agreements that might resolve it have not been signed.

Green also says retail sentiment on Nvidia cooled from bullish to neutral in a day, with chatter volumes falling from high to normal. He calls that an early indication and not a conclusion, and does not say where the data comes from.