When Expectations Become Financial Risk
The concentration and preparedness described earlier do not stop at production. They shape who borrows, whose assets rise, which households depend on the boom and how much information supervisors and policymakers have about the exposures. This chapter traces that financial and policy chain, showing why leverage and valuation risk can appear before a productivity shortfall is proven, and why monetary, fiscal and macroprudential choices must work across more than one diffusion path.
The central distinction is between a technology risk and a transition risk. A technology risk concerns whether a system works as expected. A transition risk concerns the financing, distribution and institutional adjustments that occur while people are still finding out. AI may be genuinely useful in the long run and still generate losses, tight credit or political resistance if investment and expectations move faster than cash flow, labor reallocation and public capacity.
The balance-sheet chain inside the boom
Debt-financed investment is the first link. AI firms have expanded capital expenditure on data centers and related infrastructure while free cash flow has declined. Private credit and bond markets have become important sources of funding. Borrowing allows firms to build ahead of current revenue, which can be rational when future demand is strong. It also creates fixed claims that must be serviced even if the technology, pricing model or customer base changes.
The risk is not simply that a project fails. A project can be useful and still become difficult to finance if its revenue arrives later than expected, if its equipment becomes obsolete more quickly, or if the cost of power and specialized inputs rises. The lender cares about repayment and collateral; the equity holder cares about the residual value after those claims are met. A high expected return does not eliminate the timing mismatch between present obligations and future productivity.
Capital obsolescence makes AI-related assets unusual in a way that matters for credit. A building or power connection may remain useful for years, while computing equipment and model infrastructure can face faster technical change. The IMF therefore highlights uncertainty around earnings and collateral values. If lenders cannot tell how durable an asset's cash flows are, they may tighten terms even when the underlying technology continues to improve. Financial stress can begin with uncertainty about value, not with proof that the technology has stopped working.
Valuations add an expectations channel. AI firms' rising market values can make it cheaper to issue equity, support investment and increase household wealth. Those effects can reinforce demand before productivity statistics catch up. But a market value is a claim on future earnings. If expectations rise faster than realized cash flow, repricing can reduce the value of collateral, raise the cost of capital and delay new investment. The same price increase that helped finance the boom can amplify the downturn if the expectation changes.
Concentration makes the feedback more correlated.