IMF BIS Economy: AI Debt Threatens the Financial System

IMF, BIS and the AI economy: reading beyond prices

Two institutions rarely aligned are talking about the same thing. The IMF BIS economy theme runs through the technical documents of 2026: the debt accumulated behind artificial intelligence.

The market watches equity valuations. The structural danger lies elsewhere, in the financial leverage fueling data centers, semiconductors and computing capacity.

Here is the pattern. Here are the precedents. Here is where it leads.

The precedent: fiber optics, 1998-2002

Between 1998 and 2001, U.S. telecommunications companies issued hundreds of billions of dollars in bonds. The goal was to build fiber-optic capacity for internet demand judged to be limitless.

The mechanism was direct: leverage on the future. Companies financed physical infrastructure with debt, betting that revenues would validate the spending. Demand arrived slower than projected, and cash flows fell behind.

The result is precisely documented. WorldCom filed for bankruptcy on 21 July 2002, with 107 billion dollars in assets, the largest corporate failure of its era. Global Crossing collapsed in January of the same year. Fiber capacity sat unused for years, while banks absorbed the losses.

One detail matters. Much of that debt was backed by projected revenues that never materialized. Banks lent against a future described in the prospectuses, and the future arrived different.

The 2026 context differs. The structure remains identical.

The current pattern: the debt behind AI

The International Monetary Fund has shifted its focus from equity valuations to the debt accumulated to finance the expansion of AI. The distinction is crucial.

Tobias Adrian, director of the monetary and capital markets department, observed that current earnings rarely signal a bubble, whereas rising leverage warrants close attention. His analysis ties leverage to systemic stability.

Data centers under construction on every continent. Demand for semiconductors at historic levels. Cloud companies investing hundreds of billions of dollars.

A growing share of this spending comes from debt, rather than retained earnings. Scale amplifies the problem. Private credit, which has grown beyond traditional banking channels, finances part of this infrastructure with reduced visibility for regulators.

The mechanism: why debt changes the nature of risk

Equity bubbles destroy wealth. Excessive debt threatens the stability of the financial system itself.

The difference is structural. Equity absorbs losses: when value collapses, the shareholder pays and the system holds. Debt transfers risk along the chain: when revenues fall short, default propagates to banks, private credit funds and counterparties.

AI capex today is increasingly financed with corporate bonds, private loans and off-balance-sheet structures. This alters the risk profile of the entire sector.

A decline in expected revenues hits shareholders first, then creditors, finally credit as a whole. The sequence has repeated in every cycle of leveraged overbuilding.

Why the IMF-BIS convergence matters

The Bank for International Settlements, described as the central bank of central banks, has raised concerns analogous to those of the Fund. Two institutions with distinct mandates now use the same language.

This convergence is rare. The Fund oversees the macroeconomic stability of member countries. The BIS monitors the global banking system and cross-border capital flows.

When both flag the same vulnerability, the signal exceeds the noise. Three precedents are enough to call it a pattern: the fiber of 2000, the structured credit of 2007, the tech leverage of 2026.

The market still prices the two warnings as academic commentary. It will price them as risk when the first spreads widen.

The counter-argument: does demand hold?

An opposite reading exists. Revenues from AI services are growing at a high pace, and the spending would reflect real demand rather than a speculative bet.

The thesis deserves respect. Companies with solid balance sheets finance part of the capex with their own cash, reducing debt fragility.

The weak point remains the pace. Infrastructure spending is running faster than verifiable revenues, and the gap is covered with leverage. The history of fiber teaches that real demand arrives years after the capacity is built.

A second limit concerns obsolescence. Chips lose value fast, and hardware financed over the long term ages before the debt is repaid.

My position

I state a clear thesis. AI-related debt will produce a significant credit event before revenues justify the current spending.

The reasoning rests on three pillars: the speed of spending exceeds the growth of verifiable revenues, the share of debt financing is increasing, and historical precedents show the same arc for fiber and railroads. Every leveraged, overbuilt infrastructure has produced a credit cleanup.

This exceeds the ordinary cycle. It is a regime change in the way technological infrastructure is financed.

What would change my reading. Evidence of sharply accelerating recurring revenues from AI services, combined with predominantly equity financing, would dismantle the thesis. Until then, the risk leans toward credit.

Three implications for capital

Here is what it means for those allocating capital now.

1. Family Offices and sovereign funds (36-month horizon): reduce exposure to corporate credit tied to AI capex, favor the equity of those who control the fabs.
2. Chief Risk Officers (18-month horizon): insert into VAR models a scenario of chain default on data center debt, absent from current calibrations.
3. CFOs and Investor Relations (18-month horizon): revise the AI-based growth narrative, which a credit squeeze would make fragile before investors.

Each horizon reflects the speed with which credit retreats, faster than equity.

The prediction

I formulate a verifiable prediction. Within ten months, at least one significant issuer tied to AI infrastructure will undergo a downgrade below investment grade or a default event on debt contracted for data centers.

Confidence: 60%. Horizon: 300 days. Verification: rating actions by S&P, Moody's or Fitch on the debt of data center and cloud operators.

What to watch

• Credit spreads on data center operators' bonds relative to the investment grade benchmark.
• Share of AI capex financed with debt in the quarterly reports of the major hyperscalers.
• The language of the next BIS Quarterly Review on the risk of technological leverage.

The divergence between spending and revenues always resolves. The question remains how.