Big Tech is quietly guaranteeing up to $300bn of AI infrastructure debt

Technology companies are supporting as much as $300bn of AI infrastructure borrowing through residual-value guarantees. The mechanism is simple but consequential: a special-purpose vehicle owns chips or a data centre, raises debt against the asset and receives a promise from a technology company that the equipment will retain a minimum value. The borrower gets cheaper financing; the guarantor avoids carrying the entire project as conventional debt.

Meta pioneered the structure for large campuses. Broadcom has reportedly supplied about $29bn of guarantees connected to chips for Anthropic, while Nvidia has backed some transactions for up to 25% of asset value, including support around SB Energy’s planned data centres for OpenAI. Meta’s Hyperion structure supported roughly $28bn of financing.

The accounting treatment does not make the risk disappear. If AI demand remains strong and the assets retain value, the guarantee may never be called. If a new chip generation makes older processors uneconomic, a customer fails or a partly built campus cannot be leased, the guarantor may have to cover the difference. Rating agencies therefore estimate stressed residual values and add some exposure back into adjusted leverage.

The arrangement distributes the AI build-out across private credit, banks, special vehicles and suppliers. It also makes the total obligation harder to see from any one balance sheet. A company can appear less leveraged than an infrastructure developer while still underwriting the value on which the developer’s debt depends.

Synthesis: this is the hidden counterpart to the enormous revenue backlogs reported by AI-cloud companies. Customer commitments make projects financeable; residual-value guarantees make their collateral credible; lenders supply the construction cash. All three links depend on demand arriving before the financed equipment depreciates.