AI Circular Financing, Explained With the Actual Numbers

Circular financing is not a scandal and it is not new. It is a structure. A supplier puts capital into a customer, the customer uses that capital to buy the supplier's product, and both parties report the transaction accurately in their own accounts. Nothing is concealed. What changes is the meaning of the revenue line, because the money that produced it did not come from a buyer outside the group. This piece sets out the mechanics, the numbers that matter, the historical precedents that ended badly, and an interactive map readers can edit with their own assumptions.

WHAT CIRCULAR FINANCING ACTUALLY IS

Take three participants. A chipmaker, a model laboratory, and a cloud operator. The chipmaker announces an investment in the model laboratory. The laboratory announces a multi-year compute commitment with the cloud operator. The cloud operator announces a hardware order with the chipmaker. Each announcement is true. Each is disclosed. Read individually, each looks like independent confirmation that demand for artificial intelligence infrastructure is broad and accelerating.

Read together, the three announcements may describe a single pool of capital moving around a closed loop. The chipmaker's investment funds the laboratory's compute commitment, which funds the cloud operator's hardware order, which returns to the chipmaker as revenue. If the loop closes cleanly, the group has manufactured reported growth for all three participants without a single additional outside customer.

The test is therefore mechanical rather than moral. For every dollar of announced commitment, ask one question. Did the payer obtain that dollar from someone outside the group, or from another named participant in it?

THE FOUR SHAPES IT TAKES

Equity for compute. The supplier takes a stake in the customer, and the customer's purchase obligation is sized against the investment. The supplier has effectively pre-paid for its own future revenue and holds an illiquid asset whose value depends on the same demand thesis that produced the revenue.

Vendor financing. The supplier extends credit so the customer can buy. This is the oldest form, and the one with the clearest historical record. It converts a sales problem into a credit problem, and the credit sits on the supplier's balance sheet at the moment when the customer's ability to repay is least testable.

Capacity commitments between participants. Long-dated take-or-pay contracts between members of the group. These have the useful property of being contractual, which makes them look firm, and the awkward property of being unfunded, which makes them contingent on financing that has not yet been raised.

Letters of intent. The weakest form and the one most often quoted at full face value. A letter of intent is a stated ambition with conditions attached. It belongs in a different column from a signed, funded, binding obligation, and the interactive map below keeps that distinction visible.

A letter of intent and a funded contract are both announced in the same press release font. Only one of them is money.

THE ARITHMETIC READERS CAN RUN

Three numbers describe any ring. The first is circular share: the value of flows whose payer and payee are both inside the group, divided by total flows. Below roughly a third, the structure is ordinary strategic investing. Above roughly fifty-five percent, most of the headline capital never leaves the group.

The second is outside revenue: what the participants collectively earn from customers who are not in the ring. This is the only cash that can retire commitments permanently. Compare it against the total commitment book and the answer is expressed in years. When the answer runs to decades, the book is not being serviced by demand, it is being rolled by financing.

The third is counterparty concentration. If one participant sits on both sides of most flows, the ring has a single point of failure, and every other member's disclosed backlog depends on that entity continuing to fund, supply and buy at once.

MAP THE RING YOURSELF

The tool below is the Circular Financing Map. Load a preset, edit the arrows to match your own reading of the disclosures, and watch the circular share, the ring index and the weakest-link verdict move. The full version, with per-node exposure detail, is at /calculator/circular-financing-map.

WHAT THE HISTORICAL RECORD SAYS

The telecommunications build-out at the end of the nineteen nineties ran the same structure at comparable scale. Equipment manufacturers financed the carriers that bought their switches and fibre. Reported orders grew. When the carriers could not raise the next round, the receivables on the manufacturers' balance sheets were written down, and the write-downs arrived at the same moment as the revenue decline, because they were two views of one event.

The capacity swap arrangements of the early two thousands added a second lesson. Two carriers each bought equivalent capacity from the other, and both recognised revenue. No outside customer appeared. The transactions were legal on their face and disastrous in aggregate, because they trained the market to price a demand curve that did not exist.

The parallel is not exact. Today's participants are, for the most part, cash-generative businesses with genuine outside demand, which the telecom equipment vendors' customers largely were not. That is precisely why the outside-revenue input in the map above matters more than the circular share on its own. A ring funded by profitable participants with real third-party customers can absorb a shock that a ring of pre-revenue counterparties cannot.

WHAT WOULD CHANGE THE PICTURE

Three disclosures would settle most of the argument. First, the split between binding and non-binding commitments, stated by counterparty. Second, revenue disaggregated by whether the customer is a named investor or supplier of the reporting entity. Third, the financing plan behind multi-year capacity obligations, because a contractual commitment without a funding source is an intention, not a liability that will be met.

Until those arrive, the reader has to construct the ring from public fragments. That is what this tool is for. It does not tell you the structure is unsound. It tells you how much of the headline number is the group talking to itself, and how long the outside customers would need to keep paying for the commitments to clear.

RELATED WORK

For the systemic framing, read The Circle Tightens. For what the listing documents will have to show, read Priced for a Monopoly. For coverage arithmetic on a single entity rather than a whole ring, use the AI Commitment Coverage tool. The full series sits at The Circle and the Filing. For the electricity the same build-out consumes, see how much energy AI actually uses.