AI infrastructure debt is creating new channels through which a change in market sentiment could spread beyond technology shares, according to the European Central Bank. Its November 2026 Financial Stability Review does not predict an AI crash. It maps a vulnerability: infrastructure spending is rising faster than the internal cash available to fund it, while debt and concentrated investor exposure connect more parts of the financial system to the same narrative.
The distinction is important. The ECB is assessing how losses might travel if earnings expectations or confidence change. It is not saying that all AI investment is unproductive, that every exposed institution faces the same risk, or that a repricing is certain.
The new analysis provides a useful way to understand the November 2026 Financial Stability Review: start with data-center capital spending, then follow the financing into bond markets, private credit, investment funds and bank–non-bank connections.
Why AI infrastructure debt is growing
Training and serving advanced models requires accelerators, networking equipment, buildings, cooling and electricity infrastructure. Many of those assets require commitments years before they operate. The ECB says cumulative capital expenditure by large U.S. technology companies could reach about $5.5 trillion by 2030, while acknowledging that forecasts vary widely.
Earlier phases of the buildout were funded largely from the large cash flows generated by established technology businesses. According to the review, free cash flow for most U.S. hyperscalers turned negative in 2026 as AI capital expenditure exceeded operating cash flow. That gap is increasingly being filled with external equity and debt.
Debt can match the long life of an infrastructure asset and does not automatically signal distress. The risk changes when many firms and investors depend on the same assumptions: sustained model demand, high facility utilization, adequate electricity supply, improving compute efficiency and future earnings large enough to service the financing.
Future & Fact’s AI infrastructure guide traces the physical chain from chips to power. The ECB review adds the financial chain that determines who absorbs losses when those plans are delayed, repriced or used less than expected.
How AI infrastructure debt can move from shares to bonds
The first link is familiar. If investors reduce their expectations for future AI earnings, share prices can fall. Equity is designed to absorb that volatility before creditors are affected.
The second link is newer at the current scale. Technology companies and infrastructure projects are issuing more bonds and borrowing through private markets. The ECB also points to off-balance-sheet structures backed by future revenue. As a result, lenders and bondholders have more direct exposure to the buildout.
The third link is concentration. A relatively small group of hyperscalers, chipmakers and model developers accounts for a large share of spending and market performance. Funds holding the same companies, their debt and derivatives can experience correlated losses rather than the diversification expected from different asset classes.
The fourth link is liquidity. If investors redeem fund shares during a rapid repricing, funds may need to sell assets into a falling market. Leverage and derivative exposure can magnify those sales. This is why the review says vulnerabilities are more pronounced in parts of the non-bank financial sector than in the banking sector.
What the ECB says about banks
The review does not describe euro-area banks as the center of an imminent AI credit crisis. It says their direct lending to the AI supply chain has grown but remains limited relative to the broader system, and it emphasizes strong capital, liquidity and profitability.
Data compiled through June 2026 show €24.3 billion in lending commitments from euro-area banks to 165 identified firms in the AI supply chain and their subsidiaries, including €11 billion that had not been drawn. That figure has limits: it covers the ECB’s identified sample and does not capture every exposure booked outside the euro area.
Banks can still face indirect effects. They lend to investment funds and other non-bank institutions, hold securities, provide derivatives and serve companies whose competitiveness may change as AI spreads. A shock can therefore move through market links even when the original bank loan to an AI company is small.
Why AI infrastructure debt needs off-balance-sheet scrutiny
Some data-center arrangements separate the technology company, project developer, landlord, equipment owner and lender. That can allocate construction and operating risks among specialist parties. It can also make the full obligation harder to see from one company’s conventional debt total.
The ECB highlights external financing backed by future revenues and says transparency can be limited. Readers should not treat every lease or project-finance structure as hidden debt. The useful question is narrower: who must pay if a facility opens late, operates below plan or generates less revenue than expected?
A primary-source review should therefore examine lease commitments, purchase obligations, guarantees, minimum payments and counterparties—not only the headline bond balance. Our guide to verifying AI data-center announcements applies the same principle to permits, utilities and construction milestones.
AI valuations depend on demanding assumptions
The ECB says current valuations incorporate ambitious earnings expectations. Cheaper models, faster efficiency gains, new competitors or slower customer adoption could redistribute revenue across the sector. These developments could benefit users while still disappointing a particular company’s investors.
Conversely, strong demand and productivity gains could support part of the investment case. The review acknowledges AI’s potential to raise long-term growth. Its financial-stability concern comes from the combination of uncertainty, leverage and concentration—not from a conclusion that the technology lacks value.
What readers should watch next
- Cash flow versus capital spending: whether operating cash increasingly covers the buildout or the external funding gap widens.
- Debt structure: maturities, interest costs, collateral, guarantees and whether financing sits at a parent company or project entity.
- Facility utilization: whether contracted data-center capacity becomes productive on schedule.
- Investor concentration: how much funds, insurers and private-credit vehicles depend on the same small group of issuers.
- Power and construction constraints: delays that postpone revenue while financing costs continue.
The ECB’s warning is best read as a transmission map. AI spending begins with physical infrastructure, but the financial exposure no longer ends with the technology company’s shareholders. As debt, funds and project structures take a larger role, the same shift in expectations can reach more balance sheets at once.
This article explains a central-bank risk assessment and is not investment advice.

