The five largest U.S. hyperscalers—Amazon, Meta, Alphabet, Microsoft, and Oracle—carry $969 billion in undiscounted future data center lease commitments as of year-end 2025, of which $662 billion sits entirely off reported balance sheets in leases not yet commenced. That $662 billion in shadow infrastructure obligations now equals 113% of these five companies' combined adjusted on-balance-sheet debt, meaning their hidden AI commitments exceed visible leverage by a significant margin.
This shadow borrowing emerges through structures that circumvent traditional accounting: off-balance-sheet guarantees backstopping shorter leases, securitized assets (ABS and CMBS) in the $48.7 billion range, and private credit vehicles with no bank-style capital requirements or formal resolution mechanisms. Investment-grade utility bonds for data center power generation, joint ventures like Meta's $27 billion Hyperion vehicle (80% funded by Blue Owl with debt sold to institutional investors like PIMCO), and GPU-collateralized lending to infrastructure providers extend the exposure further off public balance sheets.
For credit investors and risk managers, the fragmentation across private credit, insurers, and non-bank lenders creates opacity that complicates contagion modeling. If AI returns prove weaker than modeled, refinancing pressures or delays in monetizing these infrastructure investments could trigger guarantee activations and force rapid repricing across off-balance-sheet structures. The BIS has flagged these arrangements as potential "shock transmission channels" if appetite for AI-linked private credit reverses.