
The European Central Bank has warned that the global artificial intelligence investment boom is becoming increasingly dependent on debt and opaque financial structures, creating risks that resemble elements of both the dotcom and subprime bubbles. The central bank's November 2026 Financial Stability Review does not conclude that artificial intelligence is itself a bubble. Instead, it focuses on how unprecedented spending on data centres, advanced chips and cloud capacity is being financed as capital expenditure begins to exceed the operating cash flows of major technology companies.
The concern is not that artificial intelligence will necessarily fail to transform the economy. The warning is that even a successful technology can be accompanied by overbuilding, misallocated capital and investor losses when financing expands faster than sustainable earnings. The review estimates that large technology companies had more than $4 trillion in off-balance-sheet commitments by the second quarter of 2026, well above their recognised on-balance-sheet liabilities. Those commitments can behave like debt even though they do not appear in the same way on corporate balance sheets.
Why AI infrastructure spending is moving toward debt
The scale of the artificial intelligence buildout is testing the ability of even the largest technology companies to fund expansion from existing profits. Data centres, electricity connections, specialised chips and cooling systems require immense upfront investment. As internal cash becomes insufficient to cover every project, companies are turning to public bonds, private credit and contractual arrangements that allow outside developers to finance infrastructure against future payments. This shift makes the AI boom more closely connected to lenders and bond markets than an expansion financed mainly by equity or retained earnings.
The structure matters because debt creates fixed obligations. A company can tolerate a decline in its share price without immediately losing the ability to operate, but it must still service loans, bonds and long-term payment commitments. If demand for computing capacity grows more slowly than expected, those obligations remain. The resulting pressure can lead companies to cancel projects, sell assets or cut other investment, transmitting an industry correction into credit markets and the broader economy.
Off-balance-sheet commitments make exposure harder to see
A central concern is the use of contracts with data-centre developers and owners of advanced computing equipment. Future payments promised by large technology firms can be used by those suppliers to secure borrowing from banks, bond investors or private credit funds. The arrangement may support faster construction, but the financial obligation is not always recorded as a conventional liability by the company promising to buy the capacity. That makes the total leverage surrounding the AI ecosystem harder for investors and regulators to measure.
Opacity becomes more important when markets are confident and funding is abundant. Investors may assess each loan or contract in isolation without seeing how several commitments depend on the same optimistic assumptions about future AI revenue. If earnings disappoint, apparently separate projects can weaken together. Losses may then appear not only in technology shares but also in corporate debt, private funds, bank exposures and investment vehicles held by households and pension savers.
Circular financing can inflate expected demand
The review also highlights circular relationships within the technology sector. Cloud providers, AI laboratories, chipmakers and infrastructure owners can hold stakes in one another, provide financing to customers or enter into mutual purchasing commitments. A chip producer may help finance a buyer that uses the funds to purchase its equipment, while the resulting order supports the revenue expectations of both parties. Such arrangements are not automatically improper, but they can make genuine end-user demand difficult to separate from demand generated by financing within the same network.
Circular links can amplify both optimism and losses. During an expansion, each contract appears to validate another company's projected growth. During a downturn, one participant's inability to meet a commitment can reduce the revenue and collateral value supporting several other obligations. The risk is greater in private markets, where disclosure may be limited and valuations may adjust less frequently than prices in public markets. That delay can conceal stress until refinancing is required or a major counterparty withdraws.
The dotcom and subprime comparisons are about financing
The dotcom comparison reflects the possibility of overinvestment in infrastructure during a period of genuine technological change. The internet continued to reshape the economy after the early-2000s market crash, but many companies and projects were financed at prices that could not be justified by their eventual earnings. That episode caused substantial investor losses, yet its wider economic impact was limited in part because equity financed a large share of the expansion.
The subprime comparison concerns leverage, complexity and hidden connections. Before the global financial crisis, debt obligations were distributed through structures that appeared to diversify risk but often made ownership and exposure difficult to understand. The current AI boom is not the same market and does not involve the same assets, but a rapid move toward credit, off-balance-sheet arrangements and interconnected counterparties raises a similar question: whether investors understand who ultimately bears the loss if optimistic cash-flow assumptions fail.
How an AI correction could reach Europe
European investors can be affected directly through holdings of AI-related shares and debt, including bonds issued in euros. Institutional funds are particularly exposed, while households can participate through direct investments, pensions and pooled funds. A sharp decline in valuations could weaken confidence and reduce spending. Wider credit spreads could also raise borrowing costs for companies and governments that have no direct connection to artificial intelligence, especially if record technology issuance competes for investor demand.
Indirect effects could be larger. A severe US technology downturn could tighten global financial conditions, reduce foreign demand and weaken risk appetite. Past credit booms that ended badly produced longer and deeper output losses than equity-led corrections. The central bank says euro-area banks retain strong profitability, capital and liquidity, but vulnerabilities are more pronounced among some non-bank financial institutions where leverage and liquidity safeguards are less developed. Interconnections between funds and banks could therefore transmit stress across sectors.
The warning calls for transparency, not a retreat from AI
The practical message is that technological promise does not remove financing risk. Investors need clearer information about long-term capacity contracts, private borrowing, guarantees and cross-ownership within the AI supply chain. Lenders must test whether projects remain viable under lower utilisation, cheaper competing models or slower earnings growth. Regulators also need to track exposures that sit outside traditional bank balance sheets, because stress may emerge first in private credit or highly leveraged funds.
Artificial intelligence may still deliver large productivity gains and support new industries. The financial danger arises when every project is priced as if that success will arrive quickly and benefit current market leaders equally. By drawing lessons from earlier booms, the review is asking markets to distinguish durable innovation from financing structures that rely on uninterrupted optimism. The next phase of AI investment will be judged not only by technical capability, but by whether the assets being built can generate enough cash to support the debt behind them.
PUBLISHED BY SUYASH PACHAURI, FOUNDER & OWNER, GLOBAL BOLLYWOOD | THE HOLLYWOOD SCOPE.