BIS Bulletin maps debt structures financing AI data-centre build-out
For the chief credit officer at a large bank
The Bank for International Settlements published BIS Bulletin No 140 on Oct. 27, in which economists Mathias Drehmann and Egemen Eren map the on- and off-balance sheet debt structures financing the AI data-centre build-out.1
According to the Bulletin, a special purpose vehicle that owns a data centre earns income from its contracts, and lease commitments, guarantees and vendor backstops from a few large firms help transform the credit rating of its debt and draw in institutional capital.1 The authors say the structures obscure the final allocation of risks, and that ultimate exposures are concentrated in a few large firms, more than their balance sheets suggest.1
According to market sources cited in the Bulletin, $75 to $100 billion of bonds and asset-backed securities linked to these vehicles have been placed so far this year.1
Moody's Ratings (2026) puts the largest hyperscalers' data centre lease commitments at $1.2 trillion by mid-2026, of which around $820 billion relate to leases that have not yet started, the Bulletin says.1
The Bulletin says banks often lend to both the vehicles and the investors holding the debt, and that neoclouds and AI labs, mostly unrated or rated below investment grade, sign similar contracts to those of investment-grade hyperscalers.1
The Bulletin says questions have emerged in recent weeks about whether data centres' income will be enough to repay all debts, especially those linked to neoclouds; on AI-financing news days in 2026, spillovers to Treasury and investment-grade bond markets were small and short-lived.1
For banks, lending to a data-centre vehicle and to the investors that hold its debt can add up to exposure to the same few lessee firms, which the Bulletin says balance sheets understate. Risk teams may want to map these links across loan books, especially where neocloud-linked debt is involved.
Our take. A data-centre vehicle's rating rests on a few hyperscalers' lease promises, and banks often lend both to the vehicle and to the investors holding its bonds. Two loan files, one sponsor. Add them together before the next renewal.
Action. Credit risk: Total exposure to each hyperscaler sponsor across data-centre vehicle loans and lending to investors holding those vehicles' bonds.
Sources
- Primary sourceBIS maps the debt structures financing the AI data-centre build-out BIS