The vendor became the creditor
Nvidia is providing credit support for a data centre that will buy Nvidia chips. There are respectable reasons to do that, and there is one reason it should make you uncomfortable, and they are not mutually exclusive.
The Ohio announcement on 17 August has a structure worth stating plainly: up to $105 billion in credit support from Nvidia for an OpenAI data centre, $1.5 billion of equity into the developer, and Nvidia as exclusive supplier of the compute inside. The company that sells the product is financing the purchase of the product and taking a stake in the buyer’s landlord.
The defensible reading
Vendor financing is ordinary in capital-goods industries. Aircraft, turbines, mining equipment and telecoms infrastructure have all been sold this way, for a straightforward reason: the manufacturer understands the asset’s residual value better than a bank does. If the buyer fails, Nvidia is repossessing hardware it knows how to resell into a market it can see. A commercial lender underwriting the same deal is guessing at the depreciation curve of an accelerator generation that does not exist yet.
There is a second defensible reading. If the constraint on the buyer is access to capital rather than willingness to pay, then financing genuinely unlocks demand that was already real. That is not manufacturing demand; it is removing a friction in front of it.
The uncomfortable reading
The problem is that both readings produce identical numbers in the near term. Revenue booked against credit you extended looks exactly like revenue booked against a customer’s own cash — until the credit is tested. The distinction only becomes visible in a downturn, which is the least useful time to discover it.
This is not an allegation of anything improper. It is an observation that a particular kind of information has been removed from the market. When customers pay their own way, their willingness to buy is a signal about the value of the product. When the seller funds the purchase, that signal is degraded by an amount nobody outside the deal can estimate.
What the structure tells you about the bottleneck
The more interesting detail is where the equity went. Not into a cloud provider, not upstream into packaging or foundry capacity, but into an energy developer. That is a statement about what is actually scarce. If silicon were the binding constraint, the rational investment is upstream of the wafer. Investing in the party that secures land, power purchase agreements and interconnection queue position says the bottleneck sits well downstream of the chip.
What to watch
Three things will separate the readings. Whether subsequent deals of this size are financed the same way or revert to customer balance sheets. Whether the credit support is drawn or merely available — those are very different exposures. And whether Nvidia’s disclosures begin breaking out revenue associated with financed capacity, which is the single number that would settle the argument.
CNBC — Nvidia backing $105 billion in financing for OpenAI data center in Ohio → · Forbes — Why Nvidia Is Building Data Centers, Not Just The Chips Inside Them → · The Detroit News — Chip giant Nvidia reaches $1.5 billion deal for AI computing capacity →