Does the Demand Buy Itself?

The figure sold to you as market strength is demand. Nvidia is weighing whether to guarantee $250 billion for an OpenAI data center in Ohio and, alongside it, to finance OpenAI's purchase of $350 billion in Nvidia chips. Put that way, it is a giant betting on its customers. Put the money in the right order, and it is a seller lending the buyer the sum the buyer will use to pay the seller.

The demand buys itself. AI's web of cross-financing has passed $800 billion: the chipmaker invests in the labs, the labs spend on cloud, the cloud buys the chips back, and the same firms sit on several sides of the same table. The loops have names and exact figures: Stargate, the $500 billion joint venture with SoftBank and Oracle; a five-year, $300 billion cloud deal between OpenAI and Oracle; Oracle buying hundreds of thousands of Nvidia GPUs to deliver that capacity. Follow one arrow and you return to where you started.

The tell is on the books of the firm at the center of the circle. OpenAI is projected to lose about $14 billion in 2026 on roughly $25 billion in revenue, and it expects no profit before 2029, when revenue is meant to reach $100 billion. Bain reckons the sector will need $2 trillion in annual revenue by 2030 to pay for the infrastructure, and at the current pace it falls short by around $800 billion. A buyer losing money on this scale cannot carry the loop on its own cash. It carries it because the seller finances it.

This is the old machinery of the bill of exchange. A draft circulates as if it were money, endorsement after endorsement, and it works until someone presents it for payment; when they do, whoever holds it at that moment pays, and it is never the one who drew it. Circular financing does the same work on demand: it makes demand look like cash when it is a promise bouncing between a handful of signatures. As long as each party honors the next endorsement, the revenue is real on everyone's books. At the first presentation for payment, the revenue turns back into what it always was, an IOU.

The risk leaves the circle. Inside the ring the principals trade revenue and stakes; the real bill, if final demand disappoints, lands one loop further out. It is held by the index funds and pension plans exposed to a handful of names that now move the index on their own; it is held by the ratepayers of power grids built for loads only promised; it is held by the taxpayers behind subsidies and public guarantees for silicon. The circle keeps the revenue. The bill is left with those who never signed it.

The concession has to be made honestly, because vendor financing is not a swindle in itself. It has built genuine capital-heavy industries, from railways to telecoms; Noah Smith describes a cycle that can be virtuous, and some read it as the way demand and capacity line up in time. True, on one condition: that final demand, the demand of those who pay to use the product and not to build it, arrives in the end. A sound bill is one that someone, at the bottom of the chain, cashes for real money. Credit between the parties becomes a problem only when it is the only demand you can see.

Anyone working inside this wave should read every billion-dollar announcement by asking which pocket the money leaves and which pocket it returns to. An order the seller paid for measures the seller's accounting. Real market hunger is a different thing, and it does not show up on that line. Your budget, your job, your project live on final demand, not on the circular kind, and telling the two apart is the first engineering skill to have in a year of inflated figures. The machine stays one of the keys to more human work right here, when you use it to take the mechanism apart instead of admiring it: and the mechanism says to follow the money until it leaves the circle, then to look hard at who is waiting for it at the exit.

A demand that buys itself is not a demand. It is a bill of exchange, and sooner or later someone presents it for payment. And is that the day the crisis begins?