A chipmaker invests in a model developer. The model developer signs a compute contract with a cloud provider. The cloud provider buys chips from the chipmaker. Every leg is a real transaction with real cash, and the loop makes end demand look larger than the number of independent buyers would suggest.
This article is for informational and educational purposes only. It is not financial advice, investment advice, or a recommendation to buy, sell, or hold any security, cryptocurrency, or financial product. Always verify data with official sources before making financial decisions.
Short answer: what makes a deal circular
Short answer: A deal is circular when a supplier provides capital to a customer who uses it, directly or indirectly, to buy from that supplier. The revenue is real and the accounting is generally legitimate. What is compromised is the signal — revenue growth normally indicates independent demand, and in a circular arrangement some portion of it reflects the supplier’s own capital returning as sales.
Mapping the loops
The AI supply chain has four layers and money moves between them in more than one direction.
| Layer | Role | Circular exposure |
|---|---|---|
| Chipmakers | Supply accelerators | Equity stakes in model labs and neoclouds |
| Model developers | Build and serve models | Funded partly by suppliers; commit to long-term compute |
| Hyperscalers | Operate large-scale infrastructure | Invest in model labs that then buy their cloud capacity |
| Neoclouds | Specialised GPU capacity providers | Purchase chips using debt secured against those chips |
Any individual arrangement is defensible. Strategic investment in a customer is ordinary corporate practice, and vendor financing has existed for as long as capital equipment has. The concern is aggregate: when several loops operate simultaneously, apparent end demand can substantially exceed demand from parties spending money they did not receive from within the chain.
Vendor equity stakes and take-or-pay contracts
Two structures carry most of the weight.
Equity investment in a customer gives the supplier upside in the customer’s success while providing the customer capital that frequently funds purchases from that supplier. The supplier books revenue; the investment sits on the balance sheet at a valuation typically set by the same funding round the supplier participated in.
Take-or-pay compute contracts commit a customer to pay for capacity whether or not it is used. These convert into contracted backlog and remaining performance obligations, which are disclosed and widely cited as evidence of demand. The obligation is only as good as the counterparty’s ability to pay, and where the counterparty is a company funded largely by the same ecosystem, the backlog is less independent than the headline number implies.
Why circularity can inflate apparent demand
Consider a simplified chain: a supplier invests $10 billion in a customer, and the customer spends $8 billion of it on the supplier’s products. The supplier reports $8 billion of revenue. Consolidated across the ecosystem, roughly $8 billion of the apparent demand originated as the supplier’s own capital.
Nothing here is improper. Both transactions occurred, the products were delivered, the accounting follows the rules. But an analyst treating that $8 billion as evidence of independent market demand is drawing a conclusion the structure does not support. The question that separates durable revenue from recycled capital is whether the end customer generates enough external cash flow to fund purchases without further injections from within the chain.
Debt-funded GPU purchases and where that debt sits
The most leveraged part of the structure sits with specialised capacity providers financing hardware with debt secured against the hardware itself.
The collateral question is the interesting one. A GPU’s value as security depends on its resale value over the loan term, which depends on how quickly newer generations arrive and how durable demand for the previous generation proves. Lenders underwrite an assumed useful life; if that assumption is wrong, the collateral is worth less than the loan against it.
This debt frequently sits with private companies and in structures that do not appear in the public financials of the chipmakers or hyperscalers whose fortunes it supports — which means the leverage in the ecosystem is harder to observe than the revenue it produces.
The honest counterargument
Strategic investment in customers is not new and is not inherently a warning sign. Telecoms vendors financed carriers. Equipment makers have long financed buyers. In genuinely new markets, suppliers frequently must fund early customers because no independent capital is available yet, and some of those investments produced enormous returns.
The AI-specific case is also not uniform. Some end demand is unambiguously external: enterprises paying for products out of operating budgets, consumers paying subscriptions. Circularity affects part of the ecosystem, not all of it, and the analytically useful exercise is separating the two rather than labelling the whole structure.
Risks, uncertainty, and limits
Specific arrangements between named companies change frequently and the details are disclosed unevenly — some in filings, some in press releases, some not at all. This article describes structures rather than making claims about any particular company’s arrangements.
Nothing here asserts that circular financing indicates a bubble or predicts any outcome. It identifies why revenue growth in this ecosystem is a weaker signal of independent demand than it would be elsewhere.