In a recent address to the global financial community, the head of the Bank for International Settlements (BIS) sounded an alarm about the rapidly accelerating flow of capital into artificial intelligence (AI) projects. According to the BIS chief, the current surge in AI‑related capital expenditures is being financed largely through opaque forms of debt, a practice that could sow the seeds of a systemic crisis if left unchecked. The warning is rooted in a historical perspective: past episodes of speculative fervor—such as the 19th‑century railway boom and the late‑1990s dot‑com bubble—share a common thread.
In each case, investors poured money into emerging technologies or infrastructure not because of proven profitability, but because of a collective belief that the sector would reshape the economy. When the promised returns failed to materialise, the resulting corrections were severe, leading to widespread financial distress and, in some instances, lasting damage to the broader economy. The BIS chief’s concerns focus on three interrelated dimensions of the AI spending surge.
First, the sheer magnitude of capital outlays is unprecedented. Companies across sectors—from cloud service providers to automotive manufacturers—are allocating billions of dollars to develop, train, and deploy advanced machine‑learning models.
This capital rush is not limited to established firms; a proliferation of start‑ups, venture‑backed funds, and even sovereign wealth entities are joining the fray, each eager to secure a foothold in what is widely perceived as the next transformative wave of technology. Second, the financing mechanisms behind these investments are often shrouded in secrecy.
Unlike traditional corporate bonds or publicly disclosed loans, many AI projects are funded through private placements, mezzanine financing, or complex structured products that do not require the same level of regulatory scrutiny. This opacity makes it difficult for regulators, rating agencies, and even the firms themselves to gauge the true level of leverage embedded in the AI ecosystem. When debt is hidden or poorly documented, the risk of a sudden loss of confidence—triggered perhaps by a high‑profile AI failure or a regulatory clampdown—can lead to rapid deleveraging, liquidity squeezes, and contagion across financial markets. Third, the nature of AI assets themselves adds a layer of uncertainty.
Unlike physical infrastructure such as railways or factories, AI models are intangible, their value heavily dependent on data quality, computational power, and the skill of the engineers who maintain them. This intangibility makes it challenging to assess collateral value in the event of default. Traditional lenders rely on tangible assets to secure loans; with AI, the collateral is often a set of algorithms or data pipelines that may lose value quickly if the underlying technology becomes obsolete or if regulatory changes restrict data usage. To illustrate the potential fallout, the BIS chief referenced the railway mania of the 1840s, where speculative investments in rail construction led to a cascade of bankruptcies when projected traffic and revenue fell short of expectations.
Similarly, the dot‑com bubble of the late 1990s saw countless internet firms raise capital based on lofty growth forecasts that never materialised, culminating in a market crash that erased trillions of dollars in equity value. In both instances, the underlying debt structures—often high‑yield bonds or convertible notes—magnified the impact of the correction, spreading distress beyond the immediate sector.
Applying these lessons to today’s AI landscape, the BIS chief warned that a comparable correction could be far more disruptive because of the interconnectedness of modern financial systems. AI technologies are now embedded in critical infrastructure, financial services, healthcare, and national security. A shock to AI financing could therefore reverberate through supply chains, credit markets, and even sovereign balance sheets. The BIS chief also highlighted the role of policy makers in mitigating these risks.
Transparency is paramount: regulators should require clearer disclosure of AI‑related debt, including the terms, maturities, and covenants attached to financing arrangements. Standardised reporting frameworks could help investors and supervisors assess the aggregate exposure of the financial system to AI projects.
Moreover, stress‑testing exercises that incorporate scenarios of AI performance failure, regulatory intervention, or sudden shifts in data privacy laws would provide valuable insight into the resilience of banks and non‑bank financial institutions. In addition to regulatory measures, the BIS chief called on corporate leaders to adopt prudent capital‑allocation practices. Firms should align AI spending with realistic revenue projections and maintain sufficient equity buffers to absorb potential setbacks.
Diversifying funding sources and avoiding over‑reliance on short‑term, high‑cost debt can also reduce vulnerability. By treating AI investments with the same rigor applied to traditional capital projects—conducting thorough cost‑benefit analyses, evaluating return on investment, and maintaining disciplined governance—companies can safeguard both their own financial health and the stability of the broader economy. The speech concluded with a forward‑looking perspective: while AI holds immense promise for productivity gains, innovation, and societal benefit, its financing must be anchored in transparency and sound risk management.
The BIS chief urged the global community to learn from past bubbles, to recognise the perils of hype‑driven spending, and to build a financial architecture that supports sustainable AI development without compromising systemic resilience. By confronting the hidden debt and ensuring that capital flows are matched to genuine, profit‑generating AI initiatives, policymakers and market participants can help steer the AI revolution toward a stable and inclusive future.