The AI buildout has outgrown its own cash flow

Where cash flow ends, credit begins

The numbers on their own describe the size of a gap. The five largest big tech companies are set to spend over a trillion dollars on AI-related capital expenditure between 2025 and 2026, and the capital expenditure of the largest firms is now outpacing their cash flows. The difference is prompting a growing reliance on debt and, increasingly, on private credit.^[1]

This is the point where an accounting identity turns into a behavioural question. If a data centre is built with a loan rather than out of earnings, the purchasing power behind that spending is created on the lender's balance sheet: the bank or the credit fund makes the loan first, and the deposit follows. The investment then depends on lenders' appetite for risk as much as on corporate profit. An alternative reading is available: the firms may be drawing down liquid assets they already hold rather than creating new credit, in which case the same spending leaves no fresh claim in the system.^[1]

The circle that is hard to see

Much of the financing is opaque and interconnected. Under so-called circular financing, chip manufacturers and hyperscalers take equity stakes in AI firms, which in turn commit to purchasing their chips and compute, linking these players in ways that are difficult to observe and, at times, difficult to value.^[1]

In August I looked at data-centre bonds on this page and wrote that the obligation stays with a strongly rated issuer while the yield is set at speculative levels. That observation was about individual pieces of paper; what is added now is the aggregate. When the place the obligation sits and the risk its buyer is paid for come apart, the same investment looks solid at one end of the chain and fragile at the other. The circular structure hides that split, because the buyer's balance sheet and the seller's sit inside the same loop.^{[1], [3]}

The route a correction would take

On the supervisory side, ESMA sets out its own measure: the disconnect between stretched technology valuations and deteriorating macro-financial conditions increases the risk of sudden market corrections. That reading on the valuation side and the reading on the capital-expenditure side are two ends of one chain. Valuations rest on ambitious expectations of future earnings and are concentrated among a small number of firms, and because households now hold more wealth in equities a sharp repricing could pass through more forcefully to consumption. A correction would not stop in portfolios; it would reach demand through what households expect to be able to spend. There is an alternative: if equity wealth is concentrated at the top of the income distribution, the same loss of value reaches a narrow group rather than consumption broadly.^{[1], [2]}

That leaves an observable signal. If the capital expenditure of the largest firms keeps outpacing their cash flows, the share of AI investment financed by debt and private credit can be expected to rise, and that share is the quantity to watch. Its direction separates an investment programme funded out of earnings from one that depends on creditors; in the second, a disappointing return hits not only the share price but the repayment schedule.^[1]