Why the BIS Sees Financial-Stability Risks in the AI Investment Boom

AI’s economic promise and its financial risks can coexist. That is the central point in recent remarks by Pablo Hernández de Cos, general manager of the Bank for International Settlements (BIS): a technology capable of materially improving productivity can still produce an unstable investment cycle if capital spending, borrowing and interconnected corporate commitments run ahead of sustainable returns. 17

The scale makes this a macro-financial issue

The BIS estimates that the world’s five largest technology firms will invest more than $1 trillion in AI over 2025 and 2026. At that scale, the infrastructure race is no longer just a technology-sector story: it can affect investment, financing conditions and financial markets more broadly. 1

The concern is not that every dollar of AI spending is wasteful. Rather, intense competition to secure computing capacity, data centres and related inputs can encourage companies to commit capital early, before the eventual revenue from AI services is clear. The BIS’s research characterises this as an investment race in which the financing structure itself can increase fragility. 14

Why debt and private credit raise the stakes

Hernández de Cos has said that AI investment was initially financed largely through internal resources and equity but has become more dependent on external financing, often from private-credit firms. 25

That changes the downside scenario. If revenues or returns fall short of expectations, borrowers still need to service their debt. BIS research warns that debt-financed investment can expose firms to fire sales of specialised assets in a bust—an especially important risk when assets are costly, purpose-built and not easily redeployed. 14

Private credit can widen the pool of available funding, but its growing role also means supervisors need a clear view of where the exposure ultimately sits. The core issue is not the existence of private credit; it is whether leverage, collateral and interconnected commitments can be assessed before a stress event.

What “circular financing” means—and why opacity matters

The BIS has highlighted the risk from circular stakes and other arrangements that connect chipmakers, hyperscalers, AI companies and infrastructure providers. These links can include equity investments and commitments that support one another’s expansion. 1425

Such arrangements can help fund rapid build-out, but they also make risk transmission harder to map. When the same firms are simultaneously suppliers, investors, customers and financiers, a reduction in expected AI demand can affect several balance sheets at once. The BIS warns that debt and circular stakes create financial fragility and can make a bust more likely. 14

For investors and supervisors, the practical question is straightforward: who bears the loss if anticipated AI cash flows do not arrive? Opaque structures can delay that answer, increasing the chance that a repricing becomes disorderly.

Valuations and the risk of extrapolation

High valuations are not proof of a bubble, and they do not negate AI’s potential economic value. But they do make markets more sensitive to revisions in expectations. The BIS has flagged elevated valuations and the possibility that an investment race leads to excessive commitments if firms and investors extrapolate future profits too far. 114

The historical parallel is therefore not that AI is equivalent to a past technology mania. It is that transformative technologies have, at times, been accompanied by overbuilding, leverage and financial losses before their longer-term economic benefits became clear. The relevant risk is a mismatch between the pace of capital deployment today and the timing or size of future cash flows.

Productivity gains are plausible, not automatic

The BIS continues to see significant potential for AI to raise productivity. But that outcome depends on more than powerful models: firms need to redesign processes, invest in complementary systems and diffuse the technology beyond a small group of frontier companies. 17

The broader economic payoff also depends on dependable digital and energy infrastructure, a workforce with relevant skills, and the capacity of businesses and public institutions to adopt the technology effectively. In other words, large infrastructure spending does not by itself guarantee broad productivity growth.

Employment, distribution and cyber resilience

The BIS also stresses the adjustment costs of an AI-driven economy. As more capable tools apply to more tasks and occupations, labour displacement could intensify; whether new work and increased demand offset those losses remains uncertain. 3

That makes retraining, lifelong learning and worker mobility central economic policies rather than side issues. It also raises a distributional question: productivity gains may be concentrated among firms, workers and countries that already have better access to capital, compute, data and skills.

AI creates additional operational and cyber concerns. Hernández de Cos has noted that AI can amplify model risk, correlated behaviour and opaque decision chains in finance. 18 Wider dependence on a relatively small set of cloud, model and chip providers may also create common points of operational failure. Effective governance, resilience testing and meaningful human oversight therefore matter alongside financial-risk monitoring.

What policymakers and central banks should watch

The BIS case is for vigilance, not for treating AI investment as inherently harmful. Policymakers need to support the conditions that make productivity gains more likely—skills, competition, infrastructure and broad access—while monitoring leverage, valuations and hidden interconnections in the financial system. 1725

International cooperation is essential because AI supply chains, capital markets and cyber threats cross borders. The BIS frames financial stability as a global public good, making information-sharing and common analytical approaches among central banks, supervisors and regulators particularly important. 21

The next phase may change who benefits

A shift in the AI investment cycle is another source of transition risk. Goldman Sachs has described increasing enterprise deployment and a move toward inference and adoption after an initially slow start. 32 That does not imply that infrastructure spending must collapse. It does mean the eventual returns may accrue differently as AI moves from building capacity to using it.

For heavily indebted projects and late-cycle infrastructure suppliers, that distinction matters. A profitable AI economy could still produce losers if demand, pricing or the pace of further capital expenditure does not meet the assumptions embedded in today’s financing. The BIS warning is therefore best understood as a call to separate AI’s long-run potential from the financial resilience of the path used to build it. 1417