BusinessMarkets, money & strategy
ECB Financial Stability Review likens big tech's debt-funded AI build-out to the subprime and dotcom bubbles
The ECB's November Financial Stability Review warns that US big tech is increasingly financing AI infrastructure with debt, much of it off balance sheet. It likens the financing structures, not the borrowers' creditworthiness, to those seen before the subprime crisis.
From the Financial Times · AI article of November 18, 2026 and cited sources: AI capital spending has moved from being paid for out of internal cash to being paid for with borrowed money. [2] [4] [9]
SOURCES[2] ECB Financial Stability Review, November 2026 ecb.europa.eu[4] BIS Bulletin 120: Financing the AI boom bis.org[9] Securities.io: ECB Stability Review flags AI debt securities.io
AI-assisted, not human-edited. Methodology Corrections Who owns this site
- What happened
- AI capital spending has moved from being paid for out of internal cash to being paid for with borrowed money. [2] [4] [9]
- Original source
- Financial Times · AI article, independent perspective
- Brief by
- AT Pulse research desk AI-assisted; publisher and contact on the desk page
- When
- November 18, 2026 article published
- Covered
- Nov 18 edition
- Evidence
- 11 sources across 9 sites
- 1AI borrowing binge has ‘historic parallels’ with subprime…ft.com
- 2ECB Financial Stability Review, November 2026ecb.europa.eu
- 3Startup Fortune: ECB says AI borrowing boom has echoes of…startupfortune.com
- 4BIS Bulletin 120: Financing the AI boombis.org
- 5BIS Quarterly Review: On- and off-balance sheet borrowingbis.org
- 6ECB Blog: Big tech, big debt (31 Aug 2026)ecb.europa.eu
Reach: not measured yet
AI capital spending has moved from being paid for out of internal cash to being paid for with borrowed money. The ECB reports record issuance of AI-infrastructure debt and projects about $5.5tn in cumulative US big-tech capex by 2030, while noting that estimates vary widely. Coverage says capex now exceeds operating cash flow at most hyperscalers. [2] [4] [9]
Much of the exposure sits off balance sheet. Data centres are held in special-purpose vehicles or joint ventures that fund them through private placements, and the hyperscalers commit to long leases and capacity offtake contracts. Coverage of the review puts these commitments above $4tn. Other estimates are far lower, including Moody's $1.2tn. [1] [3] [5] [10]
The ECB sees risks beyond high valuations. In circular deals, suppliers invest in AI labs that then buy the suppliers' products, which may inflate projected revenue and hide real demand. Cheaper open-weight models could also erode the profits that frontier firms are counting on to service their debt. [2] [8]
Direct exposure in the euro area looks limited for now. Banks are highly profitable and lend little directly to AI. Lagarde called eurozone exposure low for the time being. However, euro-area non-banks hold 16–26% of their equity in US tech and AI firms, and credit quality in software and semiconductor loans is showing signs of weakening. [2] [7] [9]
The ECB documents that AI build-outs increasingly depend on debt and opaque lease structures.
Read the full assessment
If AI earnings disappoint, compute pricing, counterparty risk in lease and offtake contracts, and European corporate borrowing costs could all come under pressure.
Executive brief
The European Central Bank (ECB) says the largest US tech companies have more than $4tn in off-balance-sheet commitments, which is more than the liabilities on their balance sheets. It does not appear in the ECB's review overview page. The ECB's November 2026 Financial Stability Review says AI infrastructure debt issuance is at record highs. It projects $5.5tn of cumulative US big-tech capital spending (capex) by 2030. The FT reports the ECB's view that this "borrowing binge" has "historic parallels" with the subprime bubble. Coverage says the comparison is about how the financing is structured, not about whether the borrowers are creditworthy.
Read the full section
The European Central Bank (ECB) says the largest US tech companies have more than $4tn in off-balance-sheet commitments, which is more than the liabilities on their balance sheets. That figure comes from the Startup Fortune write-up of the ECB review. It does not appear in the ECB's review overview page. The ECB's November 2026 Financial Stability Review says AI infrastructure debt issuance is at record highs. It projects $5.5tn of cumulative US big-tech capital spending (capex) by 2030. The FT reports the ECB's view that this "borrowing binge" has "historic parallels" with the subprime bubble. Coverage says the comparison is about how the financing is structured, not about whether the borrowers are creditworthy.
What changed and event timeline
BIS flags the shift to debt
Bank for International Settlements (BIS) economists said AI investment would have to move from internal cash flow to debt, with private credit growing fast. They judged the stability risk "moderate" at the time ().
"Shadow borrowing" documented
The BIS described hyperscalers issuing more than $100bn in bonds in 2025. It also described data centres held in special-purpose vehicles (SPVs), with long-term leases and offtake contracts that keep the debt off the hyperscalers' balance sheets ().
Big tech enters euro bond market
An ECB blog found US big tech accounted for nearly 10% of gross new euro-denominated non-financial corporate bond issuance in 2026 ().
Lagarde letter to MEPs
ECB President Christine Lagarde warned members of the European Parliament that rising AI debt could become hard to sustain if earnings disappoint. The reporting headline said eurozone exposure "remains low for now" ().
Vujčić widens the risk lens
ECB Vice-President Boris Vujčić named borrowing, concentrated exposures, opaque off-balance-sheet structures and circular deals as risk channels beyond high valuations ().
Capabilities and access
This story is about financing, not a specific AI model or version. The documents behind it are public: The FT article is paywalled. Only its summary line was reviewed: it says hyperscalers are building up trillions of dollars in off-balance-sheet commitments through contracts with other tech groups.
Read the full section
This story is about financing, not a specific AI model or version. The documents behind it are public:
- The ECB review, including Box 3 on how AI is financed with debt and Box 4 on big tech in the euro bond market (ECB FSR).
- The BIS analyses (linked above).
The FT article is paywalled. Only its summary line was reviewed: it says hyperscalers are building up trillions of dollars in off-balance-sheet commitments through contracts with other tech groups.
Technical analysis for researchers and developers
- Funding gap: AI capex now exceeds operating cash flow.
- How the off-balance-sheet deals work: an SPV or joint venture owns the data centre and funds it through private placements.
- Circular financing: suppliers invest in AI labs that then commit to buying the suppliers' hardware or services.
Read the full section
- Funding gap: AI capex now exceeds operating cash flow. Coverage says free cash flow turned negative for most US hyperscalers in 2026 and is expected to stay negative through 2028 (Securities.io).
- How the off-balance-sheet deals work: an SPV or joint venture owns the data centre and funds it through private placements. The hyperscaler holds a minority stake and signs long leases or capacity offtake agreements (BIS).
- Circular financing: suppliers invest in AI labs that then commit to buying the suppliers' hardware or services. The ECB says this may inflate future revenues and hide real demand (ECB FSR).
- Methodology: the reviewed sources do not explain how the $4tn figure was calculated.
Claims and evidence
- Record AI-infrastructure debt issuance and $5.5tn capex by 2030: stated in the ECB FSR. The ECB itself says estimates "vary widely".
- More than $4tn in off-balance-sheet commitments: reported by Startup Fortune, but not visible on the ECB overview page. These are likely measured differently.
- More than $200bn of US-dollar investment-grade bonds issued by hyperscalers in 2026: per Securities.io.
Read the full section
- Record AI-infrastructure debt issuance and $5.5tn capex by 2030: stated in the ECB FSR. The ECB itself says estimates "vary widely".
- More than $4tn in off-balance-sheet commitments: reported by Startup Fortune, but not visible on the ECB overview page. Other estimates are lower: $1.65tn, or Moody's $1.2tn, cited in BGov-linked reporting. These are likely measured differently.
- More than $200bn of US-dollar investment-grade bonds issued by hyperscalers in 2026: per Securities.io.
Context and prior work
- The BIS warned about this shift in January.
- The Bank of England has also flagged AI-related debt as a risk (FStech).
- The ECB's own August blog estimated hyperscalers need more than $1tn of capex by 2028. The review's $5.5tn projection runs to 2030.
Limitations, safety and contested findings
- The subprime comparison is about financing mechanics: speed of debt build-up, opacity and circular deals.
- Their return on equity is at a ten-year high, and direct lending to AI is "rather limited", though credit quality in software and semiconductor loans shows signs of getting worse (ECB FSR).
- Lagarde described eurozone exposure as low for now (Central Banking).
Read the full section
- The subprime comparison is about financing mechanics: speed of debt build-up, opacity and circular deals. It does not say Microsoft or Alphabet are weak borrowers (Startup Fortune).
- Banks look resilient. Their return on equity is at a ten-year high, and direct lending to AI is "rather limited", though credit quality in software and semiconductor loans shows signs of getting worse (ECB FSR).
- Lagarde described eurozone exposure as low for now (Central Banking).
- The BIS called the risk "moderate" in January.
Business and practitioner implications
- Price competition: the ECB says cheaper open-weight models could undercut frontier firms' expected profits (ECB FSR).
- Investor concentration: US tech and AI firms make up 16–26% of euro-area non-banks' equity holdings (Securities.io).
- Contract scrutiny: lease and offtake commitments are where the hidden exposure sits. Counterparties should check them before signing.
Read the full section
- Price competition: the ECB says cheaper open-weight models could undercut frontier firms' expected profits (ECB FSR). Buyers of compute should expect price competition to affect what vendors can borrow.
- Investor concentration: US tech and AI firms make up 16–26% of euro-area non-banks' equity holdings (Securities.io).
- Contract scrutiny: lease and offtake commitments are where the hidden exposure sits. Counterparties should check them before signing.
- European borrowers: big tech's euro bond issuance could push up borrowing costs for other European companies (ECB Blog).
The source trail.
Sources (11)
AI borrowing binge has ‘historic parallels’ with subprime bubble, says ECB
Feed excerpt only.
ft.com