How the Loop Forms: Vendor Financing in the AI Stack

Follow the money through the AI stack and it starts to curve back on itself. A chipmaker sells accelerators to a neocloud. The chipmaker also invests in that neocloud, or in the AI lab that rents its capacity. The lab uses the funding to buy compute from the neocloud, which uses the revenue to buy more chips from the chipmaker. The dollar that left the chipmaker comes home as a chip order. That loop is circularity, and it is now a structural feature of how AI infrastructure is financed.

Name the players first. Chipmakers design and sell accelerators. Neoclouds buy the chips and rent out compute. AI labs rent the compute to train and serve models. In a linear world these are three arms-length parties: the lab pays the neocloud, the neocloud pays the chipmaker, cash flows one direction, and each link is disciplined by an independent counterparty. Circularity appears when the parties also finance each other — when the seller of chips is also a funder of the buyer, directly or one step removed.

Why would a chipmaker fund its own customers? Vendor financing is an old, rational tactic. It pulls demand forward. If a customer cannot afford your product outright, help finance the purchase and you book the sale now instead of waiting. It also seeds an ecosystem — early backing for promising buyers can lock in a platform and a software moat. In a supply-constrained boom, a chipmaker with more demand than it can serve uses financing and allocation to pick winners and bind them to its roadmap. Every one of these motives is defensible on its own. The problem is not any single deal. The problem is the loop they form together.

The loops take a few recognizable shapes. A chipmaker takes an equity stake in a neocloud that is a major buyer of its chips. A chipmaker or cloud invests in an AI lab, and the lab commits to spend a large share of that money on the investor's compute or silicon. A neocloud signs a giant capacity contract with a lab, then uses that contract as the security to borrow the money to buy the chips that serve it — chips from a vendor that may also be an investor in the lab. Trace any of these and you find the same shape: the same dollars circulating among a small set of interdependent players, each transaction booked as independent revenue or demand.

Here is why that shape matters, stated plainly. Circular financing can manufacture the appearance of demand. When a chipmaker funds a buyer that then buys its chips, some of the chipmaker's revenue is, in economic substance, its own capital returning as a sale. The growth looks organic; part of it is recycled. An outside analyst sees booming orders and reads genuine end-demand, when a portion is the loop turning. The signal that is supposed to validate the boom — rising sales, rising bookings — is partly generated by the boom's own financing.

The second problem is that circularity hides correlation. On paper the risk is spread across three different companies in three different businesses. In reality they are tied to one underlying bet: that AI compute demand keeps growing fast enough to justify the whole edifice. If that single assumption weakens, it does not hit one player — it hits all of them at once, because they are the same wager wearing three costumes. Diversification that is only nominal is not diversification at all.

Contrast the two worlds to feel the difference. In the linear world, an AI lab's demand for compute is validated by the lab's own independent ability to raise money and generate revenue. A neocloud that overbuilds gets punished by a market of unrelated customers who simply rent elsewhere. Each link is tested by a counterparty with its own interests. In the circular world, the counterparty testing your demand is the same entity that financed it. The discipline that is supposed to keep the system honest is compromised, because the referee has a stake in the outcome.

None of this means the demand is fake. AI compute demand is real and large. Circularity is dangerous precisely because it rides on top of something genuine, amplifying it and making it impossible to tell, from the outside, how much of the reported growth is true end-demand and how much is the loop. That is the analyst's task in the next lesson: not to declare the boom hollow, but to size the loop — to work out how much of the apparent demand can stand on its own if the financing stops circulating.