Nvidia's new plan does not look, at first glance, like the work of a hardware company. According to TechCrunch, the company is building a structure to ensure its GPUs do not lose value, and is using it to convince a new class of financiers to keep lending against AI buildouts. The scale in question is $500 billion.

The problem: accelerating obsolescence

AI data centres are built with debt, and the collateral for that debt is usually the hardware itself. A financier lends on the assumption that the GPUs being bought will still hold a certain value several years later.

That assumption has come under strain. As Nvidia's own product cycle accelerates, a card bought two years ago falls in market value quickly. The hardware still works; but because the next generation does the same job on markedly less electricity, the operating cost of running the old one can make it economically obsolete.

From a lender's perspective that means the value of the collateral cannot be predicted. Unpredictable collateral means either a higher interest rate or no loan at all.

Why Nvidia cares

Nvidia does not sell a card and walk away; whether its customer can buy the next one depends on whether that customer can raise financing. If the financing channel narrows, so does demand — and demand narrows for a reason entirely unrelated to the quality of the product.

The company therefore has a direct interest in building a structure that supports the second-hand value of its own hardware. Defending the residual value of its GPUs is, in effect, defending its own sales volume.

What makes it risky

The plan creates a two-way tie. To the extent that Nvidia underwrites expectations about the future value of its product, it also takes onto its own side a share of the risk that arises if that value does not materialise.

There is also circularity. Nvidia's sales depend on the financing its customers can raise; that financing depends on expectations about the value Nvidia's hardware will hold. A break in one link of the chain can spread quickly to the others.

Why it still makes sense

Against that, the plan has a clear logic. AI infrastructure at today's scale is financed by credit markets rather than by individual companies' cash. That market closing would halt the sector's growth far faster than any technical limit.

TechCrunch's assessment runs the same way: the move is risky but brilliant, particularly for ageing hardware. Creating a market for cards that until now quietly lost their value means collateral for the lender and cheap compute for smaller players at the same time.

The real story here is this: the bottleneck in front of AI is no longer only chip production, but where the money to buy those chips will come from.

In brief

  • Scale: $500 billion.
  • Goal: protect the resale value of GPUs.
  • Audience: financiers lending against AI data centres.
  • Risk: Nvidia takes on part of the risk tied to its own product's future value.