Skip to content

Invest1 publisher3 min readPublished

Euro-area firms fund AI without new debt on spending that grows under a third as fast as America's

Euro-area firms are paying for AI without new debt, ECB survey data shows, while the five largest US tech firms carry $3 trillion in debt. That keeps European balance sheets out of a cooling AI bond market, but the compute gap with the US still has to be closed some other way.

The Investor · Invest desk

Illustration accompanying Euro-area firms fund AI without new debt on spending that grows under a third as fast as America's
Generated illustration

What happened

  • The ECB set out the findings on October 2 in a blog post titled "How firms plan to finance AI investment," drawing on its Survey on the Access to Finance of Enterprises.
  • Oxford Economics projects US corporate spending on AI hardware and infrastructure to rise 40% in real terms from 2021 to the end of 2027, against 12% for the euro area.
  • Hyperscalers and related issuers such as Nvidia sold $225 billion of bonds by mid-2026, according to S&P Global, with close to $400 billion projected for the full year.
  • Apollo's Torsten Slok reported that investor orders per dollar of hyperscaler bonds fell below two times by July, from nearly five times in February.

Compiled by The InvestorSomething wrong?How this is made

Why it matters

  • constraint Without new borrowing, euro-area firms can spend on AI only what their existing resources allow, so closing the 14-gigawatt compute shortfall Draghi projects for 2030 one firm at a time looks out of reach.
  • exposure Delays at unfinished US data centers fall on the firms and financiers behind more than $820 billion of commitments Moody's counts, a risk the ECB survey suggests euro-area firms have mostly stayed out of.
  • contradiction Cryptopolitan's summary says euro-area firms take on no new debt for AI while its text says only that they borrow less than US firms, and the two versions imply very different credit exposure for European lenders.

The ECB finding supports three readings, and each one calls for a different policy. One is prudence: euro-area firms are staying out of the leverage US tech is building. Another is handicap. Cryptopolitan's report leans this way, describing firms that already spend far less than their American counterparts as declining to close the gap by borrowing [16]. The third, which I think the evidence supports best, is scale: European AI projects may simply be small enough to pay for without a loan. Cryptopolitan's account of the ECB post does not include the survey's percentages [1].

The forecasts favour the third reading. In my view, growth of the European size can be paid for out of operating cash, and it takes a much faster build to push a company into the bond market. On Oxford Economics' numbers, US spending on AI hardware and infrastructure grows more than three times as fast as the euro area's over the same period [5].

The five largest US tech companies hold $1.35 trillion of debt on their balance sheets and $1.65 trillion off them, according to a Nikkei study cited by Fortune [4]. That puts 55% of the total out of view [2]. The hidden portion has grown roughly eightfold in four years [5]. US firms are also borrowing to fund long-term obligations for chips, servers and leases with data-center operators [7].

That borrowing now has to find buyers in a thinner bond market. If issuance reaches the projected $400 billion, hyperscalers and their peers have about $175 billion left to sell after mid-year [3]. They will be selling to buyers whose orders per bond dollar have fallen by more than half since February [4]. "We're being very selective in terms of how we invest within hyperscaler debt," Colby Stilson, head of fixed income at Brown Advisory in London, told Reuters [9].

On the shares Draghi cited in his September column, the US has more than 15 times the EU's AI compute capacity [6]. His fix is for European firms to pool their buying power into contracts large enough to finance new data centers [15]. A contract big enough to finance a building is a multi-year purchase commitment that a developer can borrow against. In structure it is close to the data-center leases US firms are funding with debt [7]. Pooling would move the borrowing off each buyer's books and onto the developer's. "Being cut off from AI, once the economy runs on it, would be more like being cut off from the US financial system. The effects would be catastrophic," Draghi wrote [14].

The scale reading is wrong if the SAFE responses show euro-area firms that want to spend more on AI but report finance as unavailable or too expensive. In that case the gap is a credit problem and the handicap reading holds. If the US build ends in the "investment bust" the Bank for International Settlements has warned of [12], Europe's restraint will look like prudence after the fact, whatever caused it.

What to watch

  • The SAFE breakdown itself: whether euro-area firms planning AI investment report finance as unavailable or too costly, which would favour the handicap reading over the scale reading.
  • Order coverage on second-half hyperscaler bond deals against the sub-two-times July level, as issuers try to reach roughly $400 billion for the year.
  • Whether European buyers or EU institutions act on Draghi's proposal to pool purchases into contracts large enough to finance new data centers.
Loading claim ledger
Loading source directory links
Loading share composer
Loading topic controls
Loading related stories