The BIS is not saying artificial intelligence is useless. That would be the wrong reading. The more precise warning is about transmission. If firms, lenders and investors price in rapid productivity gains before those gains broaden across the economy, a correction can move beyond equity portfolios. It can affect corporate capital expenditure, credit spreads, debt-service assumptions and the appetite of lenders to finance adjacent projects.
Central-bank institutions worry about those channels because they turn a sector story into a balance-sheet story. Equity investors can absorb a repricing if the losses stay inside portfolios. Credit risk is different. Debt has covenants, refinancing dates and counterparties. If expected AI cash flows support borrowing for data centres, power infrastructure or suppliers, weaker demand can travel through lenders and corporate balance sheets.
The power and infrastructure link is important because AI spending is not confined to software budgets. Data centres require land, grid connections, chips, cooling systems and long-term energy contracts. Those commitments can be financed on assumptions about future utilisation. If utilisation falls short, the strain appears in assets that look far removed from the model providers themselves.
That is what separates an ordinary technology cycle from a macro-financial story. A failed product line hurts a company. A crowded investment thesis can hurt balance sheets if too much financing depends on the same expected future cash flows. The risk grows when the boom is concentrated among a small set of firms or suppliers, because disappointment then has fewer places to dissipate.
There is a serious counter-case. AI investment may still raise productivity if adoption spreads from infrastructure spending into ordinary business processes. The BIS's own framing leaves room for that upside. The concern is sequencing: financial markets can capitalise the promise long before the productive benefits are measured in output, wages or margins.