Pension funds and insurers in the euro area are among the buyers of the debt that pays for American AI data centres, and four staff members of the European Central Bank say the habit is growing. US big tech companies account for just shy of 10 percent of gross new issuance of euro-denominated corporate bonds, in a count by ECB staff that runs to 20 August. The same authors say the trend warrants close monitoring.
Table of contents
That puts the exposure in portfolios that back pensions and insurance policies. The authors see no alarm yet. Their data stops before the October sell-off.
A tenth of the new paper
The count comes from a blog post of 31 August by four ECB staff, who say the views are theirs and not necessarily the ECB’s. It covers five US companies the authors call hyperscalers: Alphabet, Amazon, Meta, Microsoft and Oracle. The euro accounts for close to 10 percent of the outstanding stock of bonds issued by these firms.
In the euro market they are still small in stock. They make up slightly over 1 percent of the benchmark indices for euro corporate bonds, with around €40 billion outstanding. The flow of new paper is the larger story.
Why the ECB sees no alarm yet
Hyperscalers are often rated AA- or higher, while most euro area corporate issuers sit in the A to BBB range. The authors conclude that spillovers to other euro area issuers remain limited. They add that the rating assumptions may not stand the test of time and could heighten the vulnerability to mispricing of credit risk.
Read also: Investors flinch at AI spending, Brussels targets triple
A speech on 30 October by Philip Lane of the ECB looks at the other side, the borrowers. Survey data on euro area firms lead him to say that “these developments indicate a shift towards debt financing” in the sectors most affected by AI. Bank loans were used by 19 percent of firms, with no significant difference between sectors with high and low AI exposure.
The people holding the paper
Hyperscalers accounted for 15 percent of the increase in domestic euro corporate bond holdings in the year to March 2026, with strong demand from pension funds and insurers. The ECB’s authors project that hyperscalers will need more than $1 trillion in capital spending in total by 2028, a lower-bound estimate from credit analysts that equals 3 percent of current annual US GDP.
Hyperscalers’ credit spreads have started to rise, and the premium is much higher at the long end, though longer maturities explain part of it.
The pipeline is bigger than the stock
Figures from LSEG show that US markets have absorbed more than $250 billion of tech-related debt this year, up 81 percent from last year, against $39.3 billion in Europe, up about 5 percent. The two sets of figures measure different things and cannot be set against the ECB’s count.
The forecasts point up. BNP Paribas expects euro investment-grade corporate net supply to rise 83 percent to €192 billion in 2027 from €105 billion this year, with €80 billion to €90 billion of the 2027 total coming from reverse Yankee issuance, meaning US firms borrowing in euros.
Fidelity International estimates €40 billion to €60 billion of tech issuance in the pipeline. Giulio Baratta, global co-head of investment grade finance at BNP Paribas, says “everybody knows about it and everybody needs to pay attention”.
The warnings are moving closer
Asset swap spreads (the extra yield investors demand over a safe swap rate) on the euro investment-grade index widened 11 basis points, or 0.11 percentage points, in October, and the Nasdaq lost more than $2 trillion in market value that month. Alphabet sold a €9 billion euro bond in May and launched a second deal on 2 November. Lagarde said in the last week of October that hyperscaler debt issuance had contributed to rising sovereign bond yields.
ECB Vice-President Boris Vujčić named high AI stock valuations, bond market stress and losses at non-bank lenders as the main risks on 2 November, and the ECB’s Financial Stability Review is due on 18 November. At the IMF, Jason Wu, assistant director in its Global Markets Analysis Division, said that if there are questions about monetising AI, “the equity investments will deplete in value, the debt may be difficult to service”.
Read also: Brussels wants triple the data centres
What 18 November has to answer
The ECB blog says this could be “merely the start of a financing wave of unprecedented proportions”. Its data stops on 20 August, before the October sell-off, so the Financial Stability Review on 18 November is the first chance to see whether ECB staff update the picture. The sources do not say how much of this debt individual national pension funds hold, so that part of the question stays open.
BASIS OF ANALYSIS
Basis of analysis. The analysis rests on five sources: two ECB documents and three copies of Reuters reports. The blog is the main primary source, and the Lane speech was read only in part. The ECB’s issuance share runs to 20 August, and the LSEG and BNP figures use different definitions, so they are not compared with it. Estimates from investment banks on total AI-related debt could not be confirmed and are not used.
Sources
- European Central Bank: Big tech, big debt: when US tech giants tap the euro area bond market (31 August 2026) – issuance share, index weight, holdings, ratings, spreads and the authors’ warning.
- European Central Bank: AI adoption in Europe, speech by Philip Lane (30 October 2026) – survey evidence on how firms in AI-exposed sectors finance themselves.
- Reuters via WHBL: AI-related debt binge may pressure European bond markets (3 November 2026) – LSEG volumes, BNP Paribas and Fidelity forecasts, spreads, Lagarde’s remark.
- Reuters via Global Banking & Finance Review: ECB flags AI valuations, bond stress, fund losses as main risks (2 November 2026) – Vujčić’s list of risks and the date of the Financial Stability Review.
- Reuters via Business Day: IMF warns of risks from AI market correction and war (13 October 2026) – the IMF’s warning on debt service if AI monetisation disappoints.

















