Credit Support and Debt Expansion in AI Infrastructure

Abstract

Credit support can improve the quality of a debt claim while increasing the amount of risk held by its lenders. We develop a financing model with fairly priced support and examine public financing records for AI infrastructure. Under a binding loss-rate ceiling, support expands debt capacity; sufficiently large borrowing responses raise monetary expected lender loss. Convex debt costs limit this response: once the borrower reaches its preferred debt level, further protection reduces loss without expanding borrowing. A purposive register of 15 instruments across four parent groups documents financing quantities and the obligations allocating shortfalls. CoreWeave's delayed-draw loan principal increases by USD 3.972 billion in the first half of 2026, while Applied Digital replaces a USD 300 million bridge with USD 1.590 billion of project notes. These observations establish financing changes rather than identify a causal guarantee effect. Executed contracts show that acceptance, enforcement timing and refinancing change the beneficiary and scope of support. A separate exact portfolio experiment isolates common-provider dependence, holding every claim's marginal loss distribution fixed. Creditor protection should therefore be evaluated jointly with chosen debt, contractual eligibility and provider concentration. The evidence supports a contract-based analysis of financing exposure without estimating industry default probabilities or system-wide losses.

Author abstract from the research manuscript. Typography and line breaks have been normalized; the linked source may show a different manuscript version.