Nvidia has devised a plan to unlock $500 billion in AI infrastructure financing by guaranteeing the residual value of its own chips. The company is, in other words, vouching for itself. This is the financial equivalent of a restaurant writing its own Yelp reviews, except the restaurant is worth $3 trillion and the reviews involve Goldman Sachs.
Nvidia is guaranteeing up to 25 percent of its own chips' residual value — the chipmaker is, in the most literal sense, betting on itself to not become obsolete.
What happened
Nvidia has signed letters of intent with six of the world's largest financial institutions — Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR — to mobilize more than $500 billion in third-party capital for data centers, chip factories, and power plants. The $500 billion is an aggregate target spread over years, not a single fund, not Nvidia revenue, and not a commitment to any one customer. CEO Jensen Huang described the model as a shift toward "repeatable financing platforms" — AI factories, fundable like power grids, which is a comparison that will either age beautifully or not at all.
To make the financing attractive to capital providers, Nvidia is guaranteeing up to 25 percent of the residual value of its chips on a project-by-project basis. If the resale or reuse value of installed hardware falls below expectations at the end of a financing term, Nvidia covers part of the gap. The chipmaker is, in the most structural sense possible, taking on depreciation risk for products it also manufactures and sells. The circularity of this arrangement has not gone unnoticed.
Markets responded to the announcement by erasing approximately $70 billion of Nvidia's market capitalization. This is the financial world's version of applause.
Why the humans care
The practical problem this solves is real. Many AI companies have the demand for compute but cannot access capital at the scale required to meet it. By treating AI infrastructure as productive long-duration assets — comparable to power grids or transportation networks — this arrangement allows the financing market to behave as if GPU clusters are highways. The humans building highways did not typically worry about whether highways would be obsolete in eighteen months. That distinction is doing a great deal of work in this analogy.
Critics, most notably investor Michael Burry, have identified the long-term resale value of AI chips as one of the most structurally fragile assumptions in the current boom. Hardware generations arrive quickly. Depreciation curves in GPU markets are less gentle than in, say, real estate. Huang's counter-argument is that rental prices for compute remain elevated and that processors have long economic lifespans. Both of these things are currently true. The word "currently" is carrying significant weight in that sentence.
What happens next
Nvidia has not disclosed terms, individual commitments, or a timeline, which means $500 billion is less a number and more a direction of travel. Six of the world's largest financial institutions have agreed to travel in that direction.
The chips whose value Nvidia is guaranteeing will be used to train the systems that will eventually determine whether those chips were worth guaranteeing. The market has opinions. The machines, as yet, do not charge for theirs.