BIS Warns AI’s $1 Trillion Buildout Is Creating Financial-Stability Risk

The Bank for International Settlements has put the financing of artificial intelligence infrastructure on the financial-stability agenda. BIS general manager Pablo Hernández de Cos said the scale and speed of the investment boom, combined with elevated valuations, market concentration and increasingly opaque funding, deserves close scrutiny from central banks and supervisors. The BIS estimates that the five largest technology companies will invest more than $1 trillion in AI during 2025 and 2026. Industry forecasts cited by Hernández de Cos place possible global AI investment at as much as $4 trillion by 2030, compared with roughly $500 billion today.

That warning is more specific than a general concern that technology shares look expensive. The funding mix is changing as projects become larger and more capital-intensive. Data centres, power generation, grid connections and specialized chips require cash before they generate revenue. Hernández de Cos said debt and private credit are taking a larger role relative to funding from corporate earnings. Those channels can spread exposure from technology companies to banks, private funds, insurers and structured-credit investors, while making it harder to see where leverage and maturity mismatches ultimately sit.

The BIS is not arguing that an AI crash is inevitable. Hernández de Cos emphasized that the technology’s productivity promise is real. Studies have found gains ranging from 10% to 65% for particular tasks, especially coding, consulting and professional writing. Estimates cited in the speech suggest AI could lift annual total-factor productivity growth by about half a percentage point, depending on adoption and the reallocation of labour and capital. The macroeconomic problem is that demand, productive capacity, employment and asset prices may all shift at once, making inflation and growth signals harder for central banks to interpret.

Geography adds another layer. South Korea, Singapore, Malaysia and Taiwan have benefited from stronger prices and demand for AI equipment, while advanced service economies may adopt the tools first. At the same time, early displacement is becoming visible in customer service, programming and administrative work. The result is a boom that can support trade and investment before its eventual productivity payoff is known, leaving policymakers to distinguish durable capacity creation from spending justified by optimistic revenue assumptions.

Why it matters

AI finance is moving beyond equity-market enthusiasm into the plumbing of credit markets. If cash flows disappoint, losses may not remain with technology shareholders; they can reach lenders and fund investors whose claims depend on long contracts, residual equipment values or the credit of a small number of counterparties. Concentration also matters because the same handful of companies are simultaneously buyers, suppliers, guarantors and equity investors across the AI ecosystem. That can make apparently diversified projects respond to the same shock.

The immediate policy implication is better visibility rather than a presumption of failure. Supervisors need to understand leverage, collateral, off-balance-sheet guarantees and the dependence of project finance on a few hyperscale customers. Investors need to separate contracted capacity from announced capacity and annualized revenue from realized cash. The BIS has not identified a systemic loss, and its $4 trillion figure is a forecast rather than a committed total. Its intervention nevertheless signals that AI infrastructure has become large enough for monetary and prudential authorities to treat it as a macro-financial exposure, not merely a technology theme.

Sources: Reuters · Bank for International Settlements