AI Capital Raising: When Investors Are Also Customers
AI capital raising can make a supplier an investor and a customer its revenue source. The BIS's circular-finance study asks who pays when the next round stops.
AI capital raising has a question buried inside the biggest funding announcements: does the investor also sell the company what the money will buy? A chip supplier puts capital into a model developer. The developer buys chips or compute. The supplier records sales, and the developer has a bigger order book. Both transactions can be real. Neither tells an outsider how much independent customer demand sits at the far end of the chain. A new Bank for International Settlements study puts that overlap under a microscope.
The BIS bulletin published on 1 October calls an investment relationship circular when two AI firms also have a commercial supplier-customer relationship during the period studied. It examined deals and trading links between 2021 and 2025. That is a broad definition: the investment and purchase need not be signed in the same transaction. It gives us a way to ask better questions without accusing every company in the chain of manufacturing revenue.
Follow the cash beyond the funding round
Imagine a cloud provider invests in an AI company that rents its compute. The AI company receives capital and commits to buying capacity. The cloud provider acquires a stake and an important customer. If the AI company wins paying users outside that relationship, both sides may do very well. If those users do not arrive in sufficient numbers, the investment and the sales contract are exposed to the same disappointment.
This is why the headline size of a round is a poor proxy for outside validation. A round can contain several investors. Its disclosed value is the total financing, not necessarily the amount the supplier-investor contributed. The BIS paper makes that limitation explicit. It identified 972 investment relationships between AI firms in its sample, but a relationship count is not a count of self-funded purchases. Nor does it measure how much of any one company's reported revenue came from financed customers.
The harder test is what happens after the financing runs out. Can the buyer keep paying from its own users' cash, or does the next round have to keep the order book moving? Funding totals will not separate those businesses for you.
The BIS authors offer a sensible commercial reason for the arrangement. A supplier sees the customer's use of its product before a distant lender does. It can finance growth with better information, while locking in demand for costly infrastructure. In the other direction, a customer can fund a scarce supplier to protect access to an input it cannot easily replace. Neither motive is inherently suspect. The financial exposure simply deserves to be read alongside the sales contract.
Why the supplier may be the strongest bidder
AI infrastructure ties parties together for longer than a purchase order. A data centre can be configured around a particular customer's workload. If one side builds something highly specific, it becomes vulnerable to renegotiation after the money has been spent. An investment stake can keep both parties at the table when an ordinary contract might leave too much room for a hold-up.
The BIS finds the concentration of these relationships upstream. In its sample, compute and infrastructure suppliers account for most circular investment relationships. That is where inputs can be both scarce and expensive. A company with limited independent financing choices may accept capital from the party best placed to deliver the machines. I would not call that a trick. I would call it a financing structure whose incentives need to be visible.
The physical side still has to work. We have previously argued that power procurement belongs near the start of an AI infrastructure plan. A financing deal can reserve chips or cloud capacity, but it cannot energise an unconnected site. Equally, a contracted site does not prove customers will use the compute at a price that services the debt and rewards the equity. The investors buying into this chain have to underwrite both questions. One cannot be substituted for the other.
There is an attractive version of circularity: a supplier takes early risk, its customer builds a product people want, and the two companies grow as outside revenue arrives. There is a brittle version: both firms treat future sales to each other as evidence that the other firm's financing is safe. It is usually impossible to distinguish those versions from the announcement alone. You need the terms and the cash flows.
The demand signal can be louder than the demand
The BIS authors point out that a supplier who finances a customer may see some revenue growth that follows from its own investment decision. That does not mean the revenue is fictitious. It means reported demand may partly depend on capital supplied within the same chain. A lender looking at fast-growing sales should ask who funded the customer and whether the purchases can continue without a fresh injection of capital.
There is a useful historical warning in the paper, but it is not a forecast. The authors discuss telecom equipment suppliers financing network operators in the late 1990s. When the operators' end-user revenue disappointed, suppliers lost sales and suffered on the financing side. AI is a different industry with different contracts and cash-rich participants. The comparable mechanism is the double exposure, not a guaranteed repeat of the outcome.
This double exposure is the part I would mark in red on a term sheet. If a supplier holds equity in a customer and also relies on that customer's orders, one weak business can impair two assets at once. If a lender finances a vehicle backed by those orders, the same shortfall can run further through the system. The BIS also notes that private credit and special-purpose vehicles can make the web of obligations harder to see. That is a reason to map the counterparties before praising the size of the raise.
For a private company, the missing details may be unavoidable in a public announcement. For a public company, look beyond the press release to the filings: related-party transactions, customer concentration, purchase commitments, contingent guarantees and the timing of cash receipts. An announced commitment and a disbursed investment are not interchangeable. A signed capacity reservation and recurring revenue from independent users are not interchangeable either. Those differences determine how much of the demand signal has survived contact with the market.
A better test for the next raise
Start with the investor list and draw the commercial arrows. Who supplies whom? Who is paying for chips, compute, power and facilities? Which purchase commitments depend on financing arriving first? Then ask what happens if the expected end user arrives a year late. Does the buyer still pay, can it defer capacity, and does a guarantee shift the loss back to an investor who also expected to book revenue?
Ask about the time period, too. The BIS definition counts a commercial relationship at any point during its five-year observation window. That is useful for finding overlap, but it does not prove a particular investor funded a particular invoice. Do not turn a system-level study into an allegation about a named transaction. Its real value is the test it gives us: trace each investment to the revenue it is supposed to help create, then identify the customer at the end who was not financed by a supplier.
The same discipline applies to the rest of the build. In AI capital raising, a funded campus still needs a viable site, cooling and an operating power route. Our earlier piece on what sits beyond the megawatt headline covers why capacity on paper is not the whole operating picture. In capital raising, the parallel mistake is treating capital committed inside a supply chain as though it were final demand outside it.
I am interested in the deals that show their work: cash actually paid, capacity delivered, independent customers renewing and contracts that spell out who bears the loss when a buyer misses its forecast. Capital can solve a bottleneck. It can also hide one for a while. The difference only becomes clear when the money moves through the whole chain and does not have to circle back for another round.