ECB Finds Over $4 Trillion in Hidden Big Tech Debt, Echoes Subprime
The ECB counted over $4 trillion in off-balance-sheet commitments at eight tech and AI firms, warning that the shift into debt finance echoes the dotcom and subprime bubbles.
By Grace Kim
4 min read

What's News
- ECB counted over $4 trillion in off-balance-sheet commitments at eight tech firms as of Q2 2026
- The Fed's November report separately tallied over $3 trillion in hidden tech obligations
- Meta's $27 billion Hyperion data center kept debt off its books via an 80% Blue Owl Capital joint venture
- ECB used an agentic AI model on OpenAI's ChatGPT 5.6-Sol to extract figures from SEC filings
- The ECB's next Financial Stability Review is due in May
By Grace Kim
The European Central Bank has counted more than $4 trillion in off-balance-sheet commitments at eight big tech and AI companies — and it says the number recalls the dotcom and subprime bubbles.
In its twice-yearly Financial Stability Review published Wednesday, the ECB estimated that major tech firms held "total off-balance-sheet commitments of more than USD 4 trillion as of the second quarter of 2026." That figure, it wrote, sits "well in excess of their outstanding on-balance-sheet liabilities." The sample covers Alphabet, Amazon, Meta, Microsoft, Nvidia, Oracle, Broadcom and Anthropic.
"The rapid shift towards debt finance invokes historical parallels with the dotcom and subprime bubbles," the central bank wrote.
The ECB's tally landed a day after the Federal Reserve's. The Fed's own November Financial Stability Report counted more than $3 trillion in signed leases, take-or-pay contracts and residual value guarantees at seven big tech companies, none of it booked as debt. The two counts cover different companies and use different methods.
What does the ECB's 2008 comparison look like?
In a dedicated box on AI debt financing, six ECB authors list parallels with the pre-crisis years:
- Debt that isn't booked as debt. Hyperscalers sign contracts for data centers and chips owned by others; the owners then borrow against those payments from "banks, bond investors or private credit funds." The ECB wrote that "such cash flow commitments are debt-like in economic terms, but they do not appear as liabilities on the balance sheets of hyperscalers under US accounting standards."
- Buying your own demand. Cloud providers and AI labs buy chips from makers that often finance the purchases, inflating "the expected revenues of both the chipmaker and the service provider." A footnote flags "other income" at Alphabet and Amazon from revaluing Anthropic stakes, and Nvidia's stakes in CoreWeave, OpenAI and Anthropic.
- Ratings the market doubts. Hyperscaler bond spreads have widened versus same-rated corporates, which "could be an indication that markets currently view rating agencies' risk assessments as too benign." CDS spreads for several big tech firms have hit "record levels."
- Loans that outlast the chips. AI equipment lasts one to three years, per market intelligence the ECB cites; the IMF puts GPUs at about two years against seven for standard gear. Most hyperscalers have stretched server accounting lives from three years in 2020 to six in 2026.
- Leverage outside the banks. Euro area funds and other non-banks have built significant AI exposure and started buying AI-firm debt. The ECB points to "the near collapse of Situational Awareness, a US hedge fund, in July 2026."
The Meta example is concrete: its $27 billion Hyperion data center in Louisiana was built through a joint venture in which Blue Owl Capital owned 80%, keeping the debt off Meta's books. The bonds were rated A+ and yielded 6.58%.
How bad could it get?
The ECB's downside case mirrors Citigroup's 2007 moment, when the bank absorbed $49 billion of off-books structured vehicles. If earnings disappoint, "the full debt incurred for the AI buildout would gradually be recognised on balance sheets," and leverage and ratings would worsen.
ECB Vice-President Boris Vujčić wrote in the review's foreword that "circular financing is making intrinsic demand hard to gauge," and that even banks not exposed "in material terms" could "face spillovers via strong links with non-banks."
The bank also distinguishes today from 2000: the dotcom bust brought "only a shallow recession thanks to a large share of equity finance." Credit-boom busts, it notes, historically brought "a significant and prolonged contraction in real GDP," and a U.S. shock "would likely result in a significant contraction in euro area GDP as well." The authors wrote that "the rapid acceleration of debt financing raises concerns that the AI buildout might become debt-driven and unsustainable."
How did the ECB count it?
Even the central bank needed AI. Staff extracted the off-balance-sheet figures from SEC filings using "an agentic AI model developed by ECB staff," running on OpenAI's ChatGPT 5.6-Sol. They trained it against hand-compiled first-quarter totals and checked its output manually. Anthropic's figures come from its IPO filing.
For now, one item on the ECB's tracker still holds: "So far, however, hyperscalers' earnings have broadly kept up with expectations," the authors wrote. Everything else on the list depends on that staying true — and the ECB's next review, due in May if the schedule holds, will tell.
Source: fortune.com
More from Grace Kim
Show full bio
Market editor covering people, launches and day-to-day practice in entrepreneurship, startups and venture funding at Business Bearings.
2,387 articles