Major technology companies are increasingly utilizing off-balance-sheet guarantees to finance the massive build-out of artificial intelligence infrastructure, a practice that has led to an estimated $1.65 trillion in hidden debt across five US tech giants. This figure, which includes some estimates, now surpasses their on-balance-sheet debt of approximately $1.35 trillion, making it more challenging for investors to accurately assess risk. This "phantom liability" does not appear on traditional balance sheets but could materialize as significant costs during an industry downturn.

Companies like Nvidia, Meta, and Broadcom are leveraging their strong credit ratings to provide "residual value" support for debt deals. Nvidia, for example, recently announced a $500 billion financing partnership and is poised to guarantee potentially tens of billions of dollars in AI chip debt. Meta has pioneered this template, using such backstops in roughly $27 billion and $13 billion debt packages for its Hyperion and Sopaipilla data centers, respectively. Broadcom applied a similar structure in its project Big Sky, backstopping most of a $35 billion debt deal to finance custom AI chips for Anthropic, enabling investment-grade ratings and lower borrowing costs for senior debt tranches.

While this financial engineering allows companies to bolster sales and client access without taking on direct debt, it has raised concerns among bond traders and rating agencies. Analysts at CreditSights describe it as "writing a put," which is pro-cyclical and could exacerbate boom-bust cycles. Moody's Ratings has warned that a substantial increase in these contingent obligations could limit financial flexibility and affect credit profiles, even if reported leverage remains low. S&P Global Ratings now considers Broadcom's residual value support a "contingent debt-like obligation" that will be added to its adjusted debt calculations.