The Chip Vendors Are Now Financing Their Own Customers
Three of the biggest AI compute deals this month had the seller putting up the money. Here is how to tell that structure apart from real demand, before the market does it for you.
Nvidia is negotiating a financing guarantee of roughly $250 billion to back OpenAI's lease and construction debt on a 10 gigawatt data center campus in Piketon, Ohio. Read that again. The company selling the chips is offering to guarantee the debt on the building that will house them. Reuters puts the total project cost above $500 billion, with up to $350 billion more in separate chip purchase financing.
That is not a one off. It is the dominant deal structure in AI right now, and it changes what the revenue numbers mean.
## Three deals, one shape
In the last week: Nvidia put roughly $5 billion of equity into Safe Superintelligence alongside access to its Vera Rubin systems, lifting the lab's compute by about 10x. AMD is reported to be investing up to $5 billion in Anthropic as part of a long term agreement to supply tens of billions of dollars of server capacity from 2027. And a lab with no consumer presence signed a multi year, roughly $410 million agreement with AWS for priority access to high end compute.
The first two share a shape worth naming. The supplier funds the buyer, and the buyer spends the money with the supplier. Booked as revenue on one side, an investment on the other. Both sides look like growth.
This is not fraud and it is not new. Telecom vendors did it in the late 1990s, and the semiconductor industry has always run on prepayments. But it does mean one thing plainly: a meaningful slice of reported AI infrastructure demand is not independent demand. It is the vendor's own capital coming back around.
## The market started noticing this week
On July 28, Apple briefly passed a $5 trillion market capitalization while chipmakers and AI exposed names sold off. The Guardian attributed the rotation to two things: worries about AI capital spending, and competition from cheaper Chinese rivals.
Both halves matter. The capex worry is a question about whether the circular financing above converts into paying end customers. The China half is the reason to doubt it will convert at current prices. Moonshot, Z.AI and DeepSeek are now serving frontier class models, including Kimi K3, with some tokens priced under 2 percent of what Anthropic charges for Fable. US developers are picking them up in spite of export controls.
Put the two together. Western labs are financing enormous fixed compute commitments at the exact moment the price of the output is collapsing. That is the squeeze, and vendor financing does not relieve it. It postpones it and concentrates it.
## Where the honest demand signal is
If vendor funded deals are a noisy signal, what is a clean one? Money from buyers with no stake in the seller. Two examples from this month sit on the other side of that line: CXMT's Shanghai listing raised at least $8.6 billion with shares up more than 450 percent on debut, and Korean financial investors put up first tranche financing for the Namyangju X-AI data center alongside LG CNS. Public market and third party capital, priced by people with nothing to sell.
That is also why the national compute build out keeps showing up in our deal flow. Sovereign buyers are slow, but their money is genuinely exogenous.
## What to do with this
**Ask who funded the purchase order.** For any AI infrastructure number you are handed this quarter, whether in an earnings call or a fundraise deck, find out whether the customer's capital came from the vendor booking the revenue. If it did, discount it. That single question separates demand from recycled balance sheet, and almost nobody asks it out loud.
**Watch the spread between compute commitments and token prices.** Fixed multi year compute deals are being signed against output prices that fell toward Chinese levels within months. If that gap keeps widening through the next two quarters, the first defaults will not be at the labs. They will be at whoever guaranteed the buildings.