Credit Support, Financing Expansion, and Risk Allocation in AI Infrastructure

Abstract

Credit support shapes both the financing of AI infrastructure and the distribution of its downside exposure. We combine a 30-project powered-land inventory, a purposive register of 15 financing instruments across four parent groups, executed contracts and regulatory orders with a model of financing choice. Three results follow. First, support expands debt capacity under a binding expected-loss-rate constraint. When costly external equity induces use of that capacity, absolute expected lender losses increase; partial use can combine a lower loss rate with a larger monetary loss. This mechanism extends to risk budgets that rise with exposure, while a fixed monetary budget holds expected loss constant. CoreWeave’s delayed-draw term-loan principal rises USD 3.972 billion in H1 2026, and Applied Digital replaces a USD 300 million bridge with USD 1.590 billion of project notes. Second, payment priority, customer acceptance, enforcement timing and power-service conditions allocate shortfalls and financial flexibility across equipment, campuses and electricity investment. Third, common support providers increase portfolio covariance at unchanged marginal credit quality. In an exact illustrative experiment, concentration raises 99 percent expected shortfall by 58.9 percent. A heterogeneous-claim extension identifies each provider’s conditional loss exposure as the relevant concentration weight. Credit support is a financial production input and a contingent allocation rule. Financial-stability assessment must connect funded amounts, contractual activation states and the balance sheets supplying protection.

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