Who Holds the AI Debt? A Traceable-Share Methodology for AI Infrastructure Credit Exposure

Background: United States senators, the Financial Stability Board and the Bank for International Settlements each asserted during 2026 that the holders of artificial intelligence infrastructure debt cannot be identified, and that private credit, securitisation and off balance sheet structures conceal where the exposure sits. Each describes an opacity that no one has measured.

Objectives: To define a reproducible perimeter of United States AI infrastructure debt, establish what public records reveal about its holders, report the traceable share, and locate where the untraceable remainder concentrates.

Methods: We construct a purpose linked perimeter at 30 June 2026, map twenty holder channels to their disclosure regimes, and grade each into three tiers: named holder identifiable from a public filing, holder sector identifiable only, or untraceable. The holder distribution of registered corporate bonds is measured from Federal Reserve Z.1 table L.213; the rest are constructed and varied over 4,000 Monte Carlo draws. We also compute the age of the most recent public observation for every traceable dollar.

Results: The perimeter is $816 billion. Named holders are identifiable for 40.6 per cent, sector attribution for a further 25.1 per cent, and 34.4 per cent is untraceable. Two thirds of the remainder sits in private credit funds and separately managed accounts, one third with foreign holders of ordinary public bonds. Traceable holdings average 149 days stale.

Conclusion: Opacity is approximately one third, it is concentrated but not only where the official sector says, and the securitisation disclosure reform that dominates the policy debate would not move it at all.