What Circular Financing Actually Is, and When It Becomes a Problem

Circular financing describes any arrangement in which a supplier provides capital to a customer and the customer uses that capital, directly or indirectly, to purchase the supplier's products. Part of the supplier's reported revenue therefore originates on the supplier's own balance sheet rather than from external demand.

There is nothing exotic here. Capital goods industries have operated this way for well over a century, because the economics of large equipment sales frequently require it. A manufacturer of turbines, aircraft, rolling stock or switching equipment sells to buyers whose projects generate no revenue until the equipment is installed and operating. Somebody has to bridge that interval. Historically the manufacturer often did, because the manufacturer understood the asset better than any bank did and was the party most motivated to see the project completed.

The practice built the American telephone network, financed a substantial share of the railway expansion of the nineteenth century, and underwrote the commercial aircraft industry through several downturns. It is disclosed in filings, examined by auditors and modelled by credit analysts. It is not a loophole.

What has changed in the current cycle is scale and speed. The sums involved in artificial intelligence infrastructure are large enough that a small number of arrangements can move the reported growth rate of companies that sit among the largest in the world, and the arrangements are being made in months rather than over the multi year cycles that characterised earlier capital goods financing.

Almost all of the circularity in the artificial intelligence build out runs through three structures, and they carry very different risk.

The first is direct equity investment. A supplier takes a minority stake in a model laboratory or a compute provider. The capital is not earmarked, and the recipient is under no contractual obligation to spend it with the investor, although in practice the number of credible suppliers is small enough that a large share of it inevitably returns. This is the loosest form of circularity and also the hardest to measure, because nothing in the disclosure links the investment to the subsequent order.

The second is equity or credit extended explicitly in exchange for compute commitments. Here the linkage is contractual. The supplier provides capital or preferential access, the customer commits to purchase or deploy a specified quantity, and both obligations are documented. This is more visible than the first form and, precisely because it is documented, easier for an analyst to size.

The third is the take or pay capacity contract, and it is the structurally important one. A customer commits to pay for a defined quantity of compute capacity over a defined term whether or not it uses it. The contract has a name, a counterparty, a tenor and a price, which means a lender can advance money against it. This is the step at which a private commercial arrangement between two companies becomes a credit exposure sitting on somebody else's balance sheet, and it is the reason the compute lessor layer deserves the separate examination it receives in the companion piece on neocloud balance sheets.

The reason circular arrangements inflate the appearance of organic demand is not that anybody is hiding anything. It is that the two halves of the arrangement are recorded in different financial statements, and no rule requires them to be netted against each other.

Capital that a supplier invests in a customer is an investing cash outflow and appears on the balance sheet as an asset, usually as an equity stake or a note receivable. Orders subsequently received from that customer are operating revenue and appear on the income statement, where they contribute to the growth rate that analysts extrapolate and that valuation multiples are applied to.

Under current revenue recognition standards this treatment is correct. The accelerators are real, they are delivered, they are installed and they generate compute. The transaction has commercial substance, which is the legal and accounting distinction between vendor financing and round tripping. Round tripping involves transactions with no substance, constructed purely so both parties can record revenue, and it is unlawful. Nothing in the current cycle that has been publicly disclosed resembles round tripping.

The consequence is nevertheless real. An investor who reads only the income statement will conclude that demand is stronger and more diversified than it is, because the income statement does not distinguish between a dollar of revenue that arrived from an external customer and a dollar that the company effectively funded itself. The information required to make that distinction exists, but it is distributed across the cash flow statement, the related party disclosures and the customer concentration note, and it has to be assembled deliberately.

Circular financing is not inherently hazardous, and treating every strategic investment as evidence of a bubble is analytically lazy. The structure becomes dangerous when three conditions hold at the same time, and each can be assessed independently.

The first condition is that the recycled share is large. If a supplier's own capital accounts for a small fraction of the incremental revenue it reports, the arrangement is a marketing expense with better accounting treatment. If it accounts for a substantial fraction, the reported growth rate is partly a description of the supplier's own investment policy rather than of the market.

The second condition is that the end customer cannot fund purchases from operating cash flow. This is the difference between bridging and sustaining. A customer with strong cash generation that accepts vendor capital to accelerate a deployment is using financing as a timing tool. A customer whose entire purchasing capacity depends on external capital is not a customer in the durable sense, it is a project, and its ability to keep buying is a function of capital market conditions rather than of its own operations.

The third condition is that a third party has lent against the resulting contracts. This is the transmission mechanism. As long as circularity remains bilateral, a disappointment is absorbed by the two parties who chose to take the risk. Once a lender has advanced money against a take or pay contract whose ultimate payer is funded by the same supplier that signed it, the risk has been distributed to institutions that never evaluated the underlying demand thesis and in many cases do not know they are exposed to it.

Any one of these conditions on its own is ordinary. Two together warrant close attention. All three together is the configuration that produced the telecommunications equipment failures of 2001, and it is the configuration worth monitoring now.

Arguments about circular financing tend to be conducted in adjectives. They can be conducted in arithmetic instead, and the relevant figure is the amount of genuinely external cash entering the system for each dollar of revenue reported inside it.

Consider the loop in its simplest form. A supplier invests a sum in a laboratory. Some proportion of that sum returns as hardware orders. The laboratory also raises capital independently from investors who have no commercial relationship with the supplier, and spends some proportion of that on compute as well. The compute is bought by a lessor which borrows against contracted capacity to fund the purchase. End customers pay for the capacity.

Every dollar of hardware revenue in that loop is funded either by the supplier's own capital or by money that came from outside the loop entirely. The ratio between those two sources is the health of the arrangement. When external capital dominates, the recycled portion is a rounding adjustment to a real market. When the recycled portion dominates, the loop is largely describing itself, and its continuation depends on the supplier's continued willingness to fund it, which in turn depends on the supplier's own share price and cash position.

The simulator in this piece exists so that relationship can be manipulated directly rather than argued about. Set the recycled share, the independent capital, the leverage taken by the lessor and the proportion of capacity under contract, and the model reports the circularity ratio, the external cash entering per dollar of reported revenue, and which participant in the chain runs out of room first. The parameters are illustrative and the assumptions are stated openly. What it demonstrates is not a forecast, it is a structure.

The comparison most frequently invoked is the vendor financed telecommunications build out of the late 1990s, and it is a useful comparison provided the right lesson is drawn from it.

Equipment vendors extended very large volumes of credit to new network operators so those operators could purchase switching and optical equipment. Reported revenues grew rapidly. When the operators failed to generate traffic revenue on the schedule their business plans assumed, they could not service the vendor debt, the vendors wrote off receivables that had already been recognised as revenue, and several of the largest equipment companies in the world lost the overwhelming majority of their market value within eighteen months.

The lesson usually extracted from this is that the demand thesis was wrong. It was not. Internet traffic grew far beyond what the most aggressive 1999 forecasts projected, and the fibre laid in that period carries a large share of global traffic today. The thesis was correct and the assets were useful. What failed was the schedule. The financing matured years before the revenue arrived, and no amount of eventual vindication helps a borrower who has to pay in 2002.

The differences in the current cycle are material and should be acknowledged. Several of the largest buyers of artificial intelligence compute are among the most cash generative companies ever to exist, and their purchases are funded from operating cash flow rather than from vendor credit. The end demand is not speculative in the way that 1999 broadband demand was speculative, because paying enterprise usage exists today at meaningful scale. Those differences make the current arrangement considerably more robust than its predecessor.

The similarity that remains is the timing question. Assets with a four year useful economic life are being financed against contracts and expectations that extend well beyond that, and the entities carrying the debt are frequently not the entities with the cash flow. That is a schedule problem of the same family, even if it is smaller in degree.

Four observable developments would move this from a structural feature worth monitoring to an active credit concern, and each is visible in public disclosure rather than requiring inside knowledge.

The first is a widening gap between reported earnings and operating cash flow at the model laboratory layer, which would indicate that growth is being funded rather than earned. The second is rising customer concentration in supplier disclosures, particularly where the concentrated customers are also entities in which the supplier holds an equity position. The third is any renegotiation of a take or pay contract, because those contracts are the collateral underneath the lending layer and a single successful renegotiation reprices all of them. The fourth is a change in the terms on which lessors can raise debt against contracted capacity, which is the point at which the credit market delivers its own verdict on the quality of that collateral.

None of these has yet occurred in a way that would justify alarm. Each is disclosed, each is checkable, and each is a better use of an investor's attention than the recurring debate about whether circular financing is inherently improper.

Circular financing is a tool. It is used because large capital projects require bridging, it is disclosed because regulation requires disclosure, and it is not evidence of misconduct. It is also a mechanism that allows a capital cycle to continue past the point at which external demand would have slowed it, and that is not a moral failing either. It is simply what the structure does.

The correct posture is therefore neither dismissal nor accusation. It is measurement. Ask how much external cash is entering the system per dollar of reported revenue, ask whether the end buyers could fund their purchases without supplier capital, and ask who has lent against the contracts. Those three questions can be answered from public filings, and they will tell a reader more than any quantity of commentary about whether the arrangement is sustainable.