Set Trending Topics as a preferred source on Google.

The European Central Bank doesn’t fully trust the A.I. rush on stock markets. In its newly published Financial Stability Review, it lists A.I. valuations alongside government debt and geopolitical tensions as the biggest threats to the eurozone’s financial system. For the debt boom behind data centers and chips, it even draws parallels to the subprime crisis. Current valuations depend on “very ambitious earnings expectations for A.I.-related firms,” Boris Vujčić, the E.C.B.’s vice president, writes in the foreword. Risks to financial stability remain elevated, he said in a statement accompanying the report, partly because of “growing exposure to A.I.”

The Subprime Parallel

The E.C.B. is blunter in a separate box on the debt financing of A.I. The rapid shift toward debt “invokes historical parallels with the dotcom and subprime bubbles,” it says. The A.I. boom shares the dot-com era’s overinvestment in infrastructure, but that recession was shallow because much of the spending had been financed with equity. What recalls the subprime mortgage bubble, by contrast, is the circular financing, the off-balance-sheet deals and the opacity of private markets.

Here is how it works: Hyperscalers sign contracts committing them to use data centers and chips owned by other companies. Those owners then use the future payments as collateral to borrow from banks, bond investors or private credit funds. Economically, that is debt, but under U.S. accounting rules it does not show up as a liability on the tech giants’ balance sheets. The E.C.B. estimates these off-balance-sheet commitments of big tech at more than $4 trillion (about 3.6 trillion euros), well above their reported liabilities. The central bank arrived at that figure using an A.I. agent built on OpenAI’s GPT-5.6 Sol, which combed through filings with the Securities and Exchange Commission.

There is also a maturity problem. Hyperscalers’ bonds run longer than the chips they finance remain economically useful, which the International Monetary Fund puts at only about two years on average. Even so, the companies have extended the accounting life of their servers from three years in 2020 to six years. Markets are already showing skepticism: Spreads on tech bonds have widened relative to similarly rated companies, which, according to the E.C.B., could indicate that investors consider the credit ratings too benign. In Europe, U.S. hyperscalers already account for nearly 8 percent of investment-grade euro corporate bonds maturing in more than ten years.

One detail directly involves Anthropic: Alphabet’s and Amazon’s earnings are increasingly boosted by valuation gains on their stakes in Anthropic, booked as “other income,” the E.C.B. notes. So far, the central bank concedes, hyperscalers’ earnings have broadly kept up with expectations. But historically, credit booms that turned into busts were followed by a significant and prolonged contraction in economic output, and such a shock in the United States would hit the eurozone as well.

What Worries the E.C.B. About A.I. Stocks

A.I. remains the central theme shaping global equity markets, the central bank writes. Share prices are increasingly driven by extraordinarily strong earnings expectations and will hold up only as long as companies keep meeting them. At the same time, markets are becoming ever more concentrated in A.I. firms and A.I.-related sectors. Then there is circular financing: Because A.I. companies take stakes in one another, buy one another’s computing power and lend to one another, genuine demand is hard to gauge. Reported earnings may also be inflated by valuation gains on stakes in other A.I. firms.

The E.C.B. also sees a warning sign in initial public offerings. A.I.-related I.P.O. volumes are on track to surpass the peak of the dot-com bubble as a share of U.S. economic output. In the past, such waves have often coincided with market peaks and signaled late-cycle exuberance. The report comes one day before Anthropic’s market debut, the biggest A.I. I.P.O. in history. The E.C.B. also cites cheaper open-weight models, which could gain market share, as a risk to the profits of the industry leaders.

A.I. Debt Is Landing in Europe, Too

Bond issuance for the A.I. infrastructure buildout is hitting new record highs this year, according to the E.C.B., despite rising long-term interest rates. By 2030, capital spending by the big U.S. tech companies is projected to total $5.5 trillion (about 4.9 trillion euros). A significant share of recent A.I.-related bonds has been issued in euros, so European investors now hold more of them directly. If sentiment turns, losses could spill over from stock markets to bond markets.

Investment funds and other nonbanks are the most exposed, according to the report. They have significantly increased their A.I. exposure in recent years, and larger eurozone equity funds hold an outsize share of A.I. stocks and use more leverage through derivatives. By contrast, direct A.I.-related lending by eurozone banks “remains rather limited,” the E.C.B. says. It does, however, see early signs of deteriorating credit quality in loans to software companies and chipmakers. The bigger danger lies in the close ties between banks and nonbanks, for example through loans to hedge funds. As an example, the E.C.B. points to the near collapse in July of Situational Awareness, the A.I. hedge fund run by Leopold Aschenbrenner.

One notable aside concerns banks directly: The growing use of A.I. agents could make deposits less “sticky,” meaning easier to withdraw. That could happen, for instance, if agents automatically move savings to whichever account offers the best rate.

Energy Prices and Government Debt Add to the Risk

The A.I. risks don’t stand alone. High public debt, governments’ growing financing needs and more leveraged investors in the market for euro area government bonds raise the risk of abrupt repricing, the E.C.B. says. Rising energy prices could also weigh on the A.I. narrative, because data centers consume enormous amounts of electricity. These vulnerabilities are closely connected and could materialize at the same time. The E.C.B. considers the banks themselves resilient, thanks to strong profits, solid capital buffers and ample liquidity.

Why Analysts Are More Relaxed

Not everyone sees it so dramatically. Even the E.C.B. concedes that valuations of U.S. tech firms have recently moved well off their record highs. Morgan Stanley called the recent sell-off in tech stocks “an overreaction” and sees further volatility as a buying opportunity, while Generali Investments views the correction mainly as a repricing. And in a Bank of America survey of fund managers, which the E.C.B. itself cites, 23 percent named an A.I. bubble as the biggest tail risk in October, while 31 percent named a disorderly rise in bond yields.


Rank My Startup: Erobere die Liga der Top Founder!