The figures in this article come from a Wall Street Journal piece dated 26 July 2026, which the Reuters agency has not been able to independently verify. Neither Nvidia nor OpenAI has publicly confirmed these amounts. These are reported discussions, not a signed deal. We treat the information for what it is, and analyse it on that basis.
There are figures that stop being figures and become questions. Two hundred and fifty billion dollars is one of them. That is the amount Nvidia is reportedly considering guaranteeing to let OpenAI lease a giant data centre. Let's look at what this deal says, carefully distinguishing what is reported from what is confirmed.
What is said to be under discussion
According to the Wall Street Journal, Nvidia is said to be in talks to provide a financial guarantee of around $250 billion, intended to help OpenAI lease a 10-gigawatt data centre. Separate discussions are said to concern chip financing of up to $350 billion.
The site is almost novelistic: Piketon, Ohio, on the grounds of a former uranium enrichment plant. The development would be led by SB Energy, SoftBank's energy subsidiary. The full campus could cost at least $500 billion to build, with electrical commissioning in phases.
To give a sense of scale: 10 gigawatts is roughly the output of several nuclear reactors, or the consumption of an entire metropolitan area. We described in our article on the compute race how AI had become a heavy industry. This deal, if it materialises, would be its most spectacular illustration.
The problem: the seller finances the buyer
Here is what makes analysts wince, and it is worth taking the time to explain.
Nvidia sells chips. OpenAI buys chips, in very large quantities. In the reported arrangement, Nvidia would act as a financial guarantor to let OpenAI build the infrastructure in which it will install... Nvidia chips.
This mechanism has a name: circular financing (or vendor financing). The supplier lends or guarantees money to its customer, who uses that money to buy the supplier's products. The supplier's revenue rises, but part of that revenue is in reality money it has made available itself.
Circular financing is not illegal and not always unhealthy: a supplier can legitimately help a strategic customer grow. The risk appears when the seller's apparent health depends on its customer's ability to repay. If OpenAI ran into trouble, Nvidia would lose both a major customer and the guaranteed money. The two companies become linked far beyond a conventional commercial relationship. Comparable mechanisms were seen in telecoms in the late 1990s, and they amplified the crash when the cycle turned.
What it says about the AI economy
This deal sheds light on a fundamental tension. Model capabilities are advancing quickly, but revenue from AI remains, relative to the scale of investment, still modest. We saw Anthropic overtake OpenAI on reported revenue with around $47 billion annualised. That is considerable for a five-year-old company. It is also a fraction of what a single data centre campus would cost to build.
Hence the need for increasingly creative financing mechanisms. When capital requirements exceed what revenue and conventional fundraising can cover, you assemble structures where suppliers, energy operators and investors guarantee each other. It is ingenious, and it is fragile.
Worth noting: Nvidia is also cited as an investor in Safe Superintelligence, Ilya Sutskever's company, further extending its presence in the equity of its own customers. The company selling the shovels during the gold rush now holds stakes in several of the mines.
The two possible readings
For intellectual honesty, both camps must be presented, because neither is absurd.
The optimistic reading. AI is an industrial transformation comparable to electrification, and infrastructure of this scale demands unusual structures. Railways and electricity were financed through equally bold mechanisms, and the bet paid off. Nvidia knows real compute demand better than anyone, and if it is guaranteeing these amounts, it is because it sees solid demand ahead.
The cautious reading. When a supplier has to finance its customers to keep sales growth going, that is often a sign that spontaneous demand is no longer enough. The fact that the guaranteed amount far exceeds the financed customer's annual revenue makes the structure highly sensitive to any slowdown. We touched on these questions in our article on the AI bubble, and this deal does nothing to ease them.
What to take away
Nothing is signed, and it is possible these amounts will change or the deal will not happen at all. But the mere fact that such discussions are taking place says something important about the moment we are in: AI's capital needs have outgrown the stage where conventional financial mechanisms suffice.
For the observer, the indicator to watch is not the size of the announcements, but the gap between committed investment and revenue actually generated. As long as that gap widens, the question of sustainability will remain open, whatever the ingenuity of the structures. And if you are looking for a sign that July 2026 was a peculiar month, consider this: there was talk of a data centre built on a former uranium enrichment plant, guaranteed by a graphics card maker. Ten years ago, that sentence would have made no sense.