BIS warns AI capex surge could amplify global downturn risks

The Bank for International Settlements has warned that the artificial intelligence investment surge could reverse into a bust with wider economic consequences. The five largest hyperscalers are expected to spend more than $1tn in aggregate on AI-related capital expenditure across 2025 and 2026, according to the BIS paper.

The scale of the build-out places AI among the largest technology-led investment waves in US history, the BIS said. Its analysis follows concerns raised in the institution’s recently published Annual Economic Report and sets out a model for how heavy spending, competitive pressure and financial links could increase fragility.

The BIS frames the current AI cycle as a race among companies seeking a small number of leading positions in a “winner-take-most” market. In that setting, firms may invest ahead of confirmed demand because securing capacity early can improve their chances of becoming dominant suppliers or infrastructure providers.

That dynamic can create more spending than the market would support under a more coordinated or efficient allocation of capital, according to the BIS. If revenues fall short of expectations, the same investment that expanded capacity can weigh on balance sheets and turn an expansion into a contraction.

Over-investment risk

The BIS paper compares the AI surge with earlier boom-bust episodes, including 19th-century canal and railway investment cycles and the internet boom of the 1990s. The institution does not argue that these episodes are identical, but uses them to show how new technologies can attract capital at a pace that later proves difficult to justify through cash flows.

Using balance-sheet and deal data, the BIS model indicates over-investment of about 1.5 times the efficient level. Under conditions in which demand is less responsive, the paper finds that excess investment could rise to around three times the efficient level.

The paper also links the size of the boom to the potential severity of a reversal. A larger period of excess spending can leave more assets, debt obligations and supplier commitments exposed if expected AI revenue does not materialise.

Financing links may spread stress

The BIS identifies early commitments funded through debt and circular financing as factors that can make a bust more likely. In a circular financing arrangement, financial support and commercial commitments can move through connected counterparties, increasing dependence among companies within the same investment chain.

Such arrangements can help firms secure capacity, customers or suppliers during a rapid build-out. They can also concentrate risk if one participant becomes unable to meet its obligations, because the financial exposure is not isolated to a single company.

A network analysis in the BIS paper found that stress at one firm could spread through chains of financial exposure to other companies. The risk, according to the institution, is that balance-sheet weakness in one part of the AI investment system could transmit to lenders, suppliers or counterparties tied to the same expansion.

The BIS warning comes as AI infrastructure spending remains a central theme for technology companies, capital markets and policymakers. The institution’s paper presents the boom as a potential source of productivity gains, but also as a cycle that could create macroeconomic costs if investment runs ahead of sustainable demand.

This story draws on original reporting from Finextra Research.