The hidden $70B “shadow credit” bill behind AI debt—and why residual-value backstops change who eats losses

Bloomberg describes bond traders grappling with roughly $70B of “phantom liabilities” at major AI companies—liabilities that “don’t appear on major AI companies’ balance sheets,” but could “materialize at the worst possible time.”

In the example Bloomberg emphasizes, the backstop is implemented through residual value support: lenders finance AI infrastructure through an intermediate structure (often a special-purpose vehicle), and if the end-user contract breaks or equipment value on resale is lower than expected, the backstop provider makes up the shortfall.

That is the investor-relevant distinction: this is not simply “downgrade risk” for the borrower. It is credit risk with a contract-defined payer that sits outside the borrower’s reported leverage.

Bloomberg’s examples connect the dots across multiple AI-infrastructure deals. The common thread is that the credit underwriting is partly dependent on what happens to equipment value at the end of a lease or in an early exit scenario.

When underwriting assumes strong resale value, but the market for secondary assets is thin or demand slows, the backstop can convert “paper support” into an actual payment obligation.

Bloomberg’s central named issuer in the residual-value/backstop story is NVIDIA.

Bloomberg says Nvidia’s “residual value” support could be “potentially tens of billions of dollars,” and frames the purpose as letting counterparties rely on Nvidia’s strong credit profile to contain customer costs. Bloomberg also references that Nvidia CEO Jensen Huang discussed Nvidia may provide residual-value support “up to 25% of an opportunity,” case by case.

Investor takeaway: in this structure, the backstop provider is not a passive witness. It is the party that can absorb the residual-value gap, turning an AI procurement advantage into a contingent credit exposure during stress.

Bloomberg’s framework implicitly redraws the supply chain for credit risk.

• Upstream support risk: hardware/platform providers that can commit to residual-value support (Bloomberg’s example highlights NVIDIA and also references residual-value support concepts used by Broadcom).
• Downstream demand and contract outcomes: the data-center owners/operators that enter the lease or infrastructure contract, including Meta Platforms.
• Financing intermediaries: capital providers and credit investors that fund the SPV financing (Bloomberg names a coalition including BlackRock, Goldman Sachs, Apollo Global Management, Blackstone, KKR, and Brookfield Asset Management).

Even when operating fundamentals appear resilient, contingent credit structures can reappear as real costs during stress. The market’s instinct to focus on balance-sheet ratios can miss the specific contractual triggers—like resale-value shortfalls and lease-exit timing.

In other words, strong current cash generation does not fully immunize the backstop provider if contract terms make it the residual-value payer.

Bloomberg’s $70B “shadow credit backstop” story is less about whether AI borrowers can pay interest in normal times and more about whether the market’s assumptions hold under stress.

In this framework, the key shift is attribution of losses. The “credit event” is engineered so that when asset value assumptions break, the contract turns into a payment. That means investors should track residual-value triggers as closely as they track coupon and maturity.

For equities, this creates a mapping from credit underwriting to corporate risk: NVIDIA is the most obvious backstop-sensitive name in Bloomberg’s central example, while upstream/backstop concepts linked to chip infrastructure and downstream data-center owners can widen the circle.