ECB Sees Subprime Parallels in AI Debt; 2.7x Dot-Com Share

The European Central Bank said on Wednesday that “The rapid shift towards debt finance invokes historical parallels with the dotcom and subprime bubbles and could lead to spillovers affecting euro area financial stability”, in a box on AI debt in its November 2026 Financial Stability Review. The Financial Times reported the warning under the headline “AI debt spree has ‘historic parallels’ with subprime bubble, says ECB”.

Our number: the ECB puts overall AI investment needs at more than 3% of US GDP a year between 2025 and 2032. Against the 1.10% of GDP that the same ECB footnote gives for the internet boom of 1996 to 2003, that is more than 2.7 times the share, by our arithmetic, and more than 1.3 times the about 2.24% it gives for the Northern Pacific Railway.

Business Pill · Capex and Depreciation Explained

An evergreen Business Engineer explainer on capital spending and depreciation. It teaches the general idea only and does not discuss this story.

The key insight: The ECB says the shift to debt in AI financing invokes historical parallels with the dotcom and subprime bubbles, and it estimates big tech’s off-balance-sheet commitments at more than USD 4 trillion as of the second quarter of 2026, well in excess of their on-balance-sheet liabilities.

What the ECB Says About AI Debt

In Box 3, ECB staff write that cumulative capital expenditure by US big tech firms “is forecast to reach about USD 5.5 trillion by 2030, although estimates vary widely”, and that free cash flows “turned negative for most US hyperscalers in 2026, as capital expenditure on AI infrastructure exceeded operating cash flows”.

The box says firms are “increasingly using opaque off-balance-sheet structures to fund the AI infrastructure buildout”. Hyperscalers commit to use data centres and chips owned by others, and those owners borrow against the expected cash flows. The ECB says such commitments “are debt-like in economic terms, but they do not appear as liabilities on the balance sheets of hyperscalers under US accounting standards”.

It estimates that big tech firms have “total off-balance-sheet commitments of more than USD 4 trillion as of the second quarter of 2026”, “well in excess of their outstanding on-balance-sheet liabilities”. The ECB says the figures were extracted from SEC filings by an agentic AI model developed by ECB staff and then cross-checked manually; its sample includes Anthropic, using its IPO filing.

Annual investment as a share of US GDP, as given by the ECB: AI investment needs more than 3% a year in 2025-2
Annual investment as a share of US GDP, as given by the ECB: AI investment needs more than 3% a year in 2025-2032, the Northern Pacific Railway about 2.24% in 1870-1890 and the internet boom 1.10% in 1996-2003. Our arithmetic: more than 2.7 times the internet boom and more than 1.3 times the railway. Source: ECB Financial Stability Review, November 2026, Box 3 (citing Van Nieuwerburgh, NBER Working Paper 35865). Not investment advice.

The Chip-Life Mismatch

The ECB writes that “the maturity of hyperscalers’ outstanding debt exceeds the economic lifetimes of a substantial share of the physical capital funded by this debt, especially in the form of advanced chips”. Its chart notes say market intelligence puts the economic life of AI equipment at one to three years, while most hyperscalers raised the accounting life of servers and network equipment from three years in 2020 to six years in 2026.

Credit spreads on hyperscaler debt have widened relative to similarly rated issuers, particularly at longer maturities, the box says, adding: “This could be an indication that markets currently view rating agencies’ risk assessments as too benign”.

The ECB also points to circular links. It says mutual finance commitments between chipmakers and their customers “inflate the expected revenues of both the chipmaker and the service provider”, and a footnote says “Other income” reported by Alphabet and Amazon comes from revaluations of private equity stakes in Anthropic.

Why Subprime, and Why Dot-Com

The box draws two separate comparisons. It says the capital spending is “reminiscent of the dotcom bubble of the early 2000s”, which saw overinvestment but “only a shallow recession thanks to a large share of equity finance”. The subprime comparison is about the funding: the ECB says the rapid acceleration of debt financing, circular financing, off-balance-sheet transactions and opaqueness in private markets are dynamics that “were evident during the subprime mortgage bubble”.

The ECB also writes: “So far, however, hyperscalers’ earnings have broadly kept up with expectations”.

The Euro Area Exposure

In a second box, the ECB says hyperscalers have issued “over USD 200 billion this year alone” in the US dollar investment-grade bond market and have branched out into euro debt, raising euro area investors’ exposure. Its foreword says banks are not exposed to a potential correction in AI-related assets “in material terms” but “could face spillovers via strong links with non-banks”.

“Risks to euro area financial stability remain elevated, given persistent geoeconomic uncertainty and growing exposure to AI,” ECB Vice-President Boris Vujčić said in the press release. Two days earlier, his Euro Finance Week speech previewed the AI risks.

The Business Lever

The lever is the capital structure of the AI build. Much of what the ECB describes appears in filings we have read this week: Nvidia’s neocloud commitments and the new guarantee line in its third-quarter CFO commentary, the lenders’ chip-curb clause on the Bridge Data Centres loan, bank-sold data centre debt and Tencent’s move to net debt on compute buying.

The Structural Read

As we read the box, the ECB separates the two comparisons: dot-com describes the scale of the spending, while subprime describes how it is funded, through off-balance-sheet commitments, circular financing and opaque private markets.

As we read it, the sharpest number is a mismatch the ECB documents: market intelligence puts the economic life of AI equipment at one to three years, while most hyperscalers raised the accounting life of servers to six years, and the ECB says their debt outlasts much of the equipment it funds.

ECB Financial Stability Review, November 2026, Box 3

“Such cash flow commitments are debt-like in economic terms, but they do not appear as liabilities on the balance sheets of hyperscalers under US accounting standards.”

Three Implications

FOR HYPERSCALERS The ECB says leverage metrics and credit ratings could deteriorate if earnings fall short, as the full debt incurred for the AI buildout would gradually be recognised on balance sheets.

FOR BOND INVESTORS The ECB says hyperscaler credit spreads have widened relative to similarly rated issuers, particularly at longer maturities.

FOR EURO AREA BANKS The ECB says banks are not exposed to a potential correction in AI-related assets in material terms but could face spillovers via links with non-banks; this is not investment advice.

The Business Engineer Lens

This story maps onto the Business Engineer framework The AI Capex Map & The State of AI Hyperscalers.

The framework says the capex race “It is a sovereign-scale financing problem dressed in tech-company clothing”.

As we read it, the ECB box gives that description a regulator’s numbers: more than USD 4 trillion of off-balance-sheet commitments and investment needs above 3% of US GDP a year. The framework predates the ECB review and does not discuss it.

What Is Not Established

The text of the box does not give a per-company split of the USD 4 trillion; that sits in its chart. The ECB calls the figure an estimate and says the economic-life range for AI equipment comes from market intelligence. The ECB does not say a crisis is coming; it describes risks and parallels.

We did not read the full Financial Times article, which is paywalled; every figure and quotation here is from the ECB’s own publication. This is not investment advice.

This Week In The AI Supercycle

AI Lost Its Central Banker, the Lab Oligopoly, and the Future of Open Source

This week’s issue: the Nvidia leadership change, read as the AI build-out losing its central banker: who now finances the GPU landlords, backstops the data centres and sponsors open models, and why two labs are becoming the bank.

Read the full issue →

Business Engineer Framework

The AI Capex Map & The State of AI Hyperscalers

The Business Engineer analysis of how the AI capex race is financed, and who is exposed when markets reprice it.

Read the Map of AI →

The Bottom Line

The ECB says the shift to debt in AI financing invokes parallels with the dotcom and subprime bubbles, and estimates big tech’s off-balance-sheet commitments at more than USD 4 trillion. Its own comparison puts AI investment needs at more than 3% of US GDP a year, more than 2.7 times the internet boom’s share, by our arithmetic.

97,000+ executives read Business Engineer for the AI strategy frameworks cited by ChatGPT, Claude, and Perplexity.

A note on sourcing. All figures and quotations are from the European Central Bank’s Financial Stability Review of November 2026 (Box 3, Box 4, the foreword) and its press release of 18 November 2026. The Financial Times headline is quoted as published; we did not read the paywalled article. The multiples of the internet boom and railway shares of GDP are our arithmetic from the ECB’s own footnote figures. This is not investment advice. We did not contact the ECB.

Sources: ECB Financial Stability Review, November 2026, Box 3 · ECB Financial Stability Review, November 2026 · ECB press release, 18 Nov 2026

Scroll to Top
FourWeekMBA