News

AI borrowing binge has ‘historic parallels’ with subprime bubble, says ECB

The ECB's Financial Stability Review, published on Wednesday, warns that the debt-fuelled global AI investment boom has 'historical parallels with the dotcom and subprime bubbles'. It estimates that big tech firms had more than $4tn of off-balance-sheet commitments as of the second quarter of 2026, 'well in excess' of their on-balance-sheet liabilities. These come from contracts with 'data centre developers and advanced chip owners': the promised future cash flows are used to secure borrowing from banks, bond investors and private credit funds, yet the commitments, economically similar to debt, are not recognised as liabilities on hyperscalers' balance sheets. The ECB also flags circular deals, in which cloud providers and AI labs borrow from chipmakers to buy their products, which 'inflate the expected revenues' of both sides, and the 'opaqueness' of AI-related debt. 'Even if AI as a technology successfully transforms the global economy, the credit-driven AI boom may result in overcapacity, capital misallocation and credit losses for investors,' it says. Surging US tech bond issuance could in future crowd out governments and other issuers, though the ECB sees little evidence of that yet. Separately, it warns of 'abrupt repricing episodes' in euro-area sovereign bonds as French political instability builds ahead of April's presidential election: French long-term yields briefly topped 5% last month for the first time in more than 20 years and the French-German spread reached 1.75 percentage points, the highest since 2011, but the ECB sees no case for its crisis tool for now. The article names no companies and gives no euro-area bank exposure figures; all quotes come from the report.

Why it matters

A central bank now puts a number on the hidden leverage behind the AI build-out: more than $4tn of off-balance-sheet commitments, above big tech's on-balance-sheet liabilities. That is the debt that ultimately funds the data centres, chips and AI labs whose rounds and valuations Dealroom tracks, and the ECB's warning of credit losses 'even if AI succeeds' is a direct challenge to the circular-financing model. It follows yesterday's row on banks curbing their AI debt exposure, now with an official regulator's view behind it.

Executive takeaways

1) The ECB's Financial Stability Review compares the credit-driven AI boom to the dotcom and subprime bubbles, nearly two decades after the subprime crisis. 2) Its key figure: big tech's off-balance-sheet commitments exceeded $4tn at Q2 2026, 'well in excess' of on-balance-sheet liabilities; this is an ECB estimate. 3) The mechanism: contracts with data-centre developers and chip owners create promised cash flows that are used to raise bank, bond and private-credit debt, but don't show up as liabilities for hyperscalers. 4) Circular deals between chipmakers, clouds and AI labs 'inflate the expected revenues' of both sides and add opacity. 5) Even if AI succeeds, the ECB sees risks of overcapacity, capital misallocation and credit losses; crowding out by tech bond issuance is possible, but there is little evidence of it so far.

What Financial Times may be missing

The article names no companies, projects or lenders, so it is unclear which hyperscalers, data-centre developers or chip financings make up the >$4tn, or how it is calculated. It gives no figures on euro-area banks', insurers' or funds' exposure to this debt, and no policy recommendations (disclosure, capital or accounting changes). No ECB official is quoted beyond the report. It also doesn't separate firm, contracted commitments from optional or cancellable ones, which matters for how close these really are to debt. On the other side, the ECB itself says markets are functioning smoothly and crowding out isn't visible yet, so the warning is about vulnerability rather than stress already under way.

Read the full article: Financial Times

More top stories