Alan Greenspan died on 22 June, aged 100. Six days later, the Bank for International Settlements published the closest equivalent this cycle has produced to his most famous warning. On 5 December 1996, Greenspan asked how we would know when "irrational exuberance" had unduly escalated asset values; the S&P 500 closed near 740 that week. Over the following 39 months, it more than doubled before the bubble burst in March 2000. His warning was analytically prescient and practically useless: anyone who de-risked on it endured three years of severe underperformance before eventually being proven right.

The BIS Annual Economic Report of 28 June warns that the artificial intelligence investment boom could end in a protracted bust with knock-on effects across financial conditions. The five largest hyperscalers are set to spend more than $1tn on AI-related capital expenditure across 2025 and 2026, commitments that now outpace their earnings and free cash flow. Consensus estimates compiled by Bloomberg show the five firms' aggregate free cash flow collapsing from roughly $395bn in 2024 to about $50bn in 2026, even as free cash flow per share for the S&P 500 keeps climbing. The BIS supplied its own historical parallels, canals, railways, electrification, the dot-com surge, each a genuine breakthrough financed beyond what returns could justify, each ending in an investment reversal.

Chart 1: Free cash flow of the five hyperscalers, USD bn - actual and consensus estimates

Source: Syz, Bloomberg

In a debt-financed capex bust, credit and order books deteriorate before equity indices. That ordering is the design principle for anyone determined not to be caught, and every indicator that follows is observable with standard market data.

Start with credit, because it is already sending signals. A sustained widening in AI-linked CDS spreads, especially if it broadens across the sector, would be the first warning that financing conditions are deteriorating before equities fully adjust. Listed business development companies offer a daily window into the private credit channel. If their discounts to net asset value widen, or payment-in-kind income rises, it would suggest borrowers are under pressure before default rates start to move.

2. Demand is the key stress test

The next step is to test demand, because capex is only sustainable if revenues keep following. The clearest warning would be a divergence between cloud revenue growth and capex growth: if cloud revenues slow while capex continues to accelerate, supply commitments are starting to run ahead of demand.

Earnings calls should also be monitored for a change in language. Words such as “digestion”, “optimisation” or “discipline” around capex guidance would suggest that management teams are preparing investors for slower spending. Finally, watch earnings revision breadth. If revisions for the AI complex start to weaken relative to the rest of the market while capex plans remain high, margins would be absorbing the gap between spending and revenues.

Finally, use market internals to confirm what fundamentals diagnose. Semiconductors have led every major capex-cycle turn in modern market history. A sustained downtrend in the SOX index relative to the S&P 500 would be one of the clearest technical warnings that the AI capex cycle is losing momentum.

Breadth provides the second confirmation. A rally supported by a shrinking share of NYSE members is a classic late-cycle warning. In 1998-2000, breadth peaked roughly 18 months before the index itself.

No single signal is enough, but a credit warning combined with weaker demand would mark the regime change. Market internals would provide the timing trigger.

The most important message from the indicators above is that none of the decisive signals has fired. Cloud revenues continue to grow, earnings revision breadth remains positive, and the credit widening identified by the BIS is selective rather than systemic. Nothing in the aggregate data points to an imminent end to the AI investment cycle.

The BIS is not giving investors a sell signal or a timetable. It is giving them a framework for what to watch before the cycle turns. The rational position is therefore engaged vigilance, not retreat, stay with the theme while the dashboard remains healthy, and watch the financing rather than the headlines.

Greenspan asked whether prices had risen too far. The BIS is asking whether the financing structure behind the AI boom can remain intact. The valuation debate is noise; the financing data is signal.