Nvidia, Broadcom Face Hidden $70B Liabilities From AI Chip Financing

The emergence of residual value guarantees reflects the tension between explosive demand for AI chips and the fragile financial structures being used to fund that buildout. Nvidia and Broadcom are effectively using their strong credit ratings as a subsidy for AI companies' infrastructure purchases, while keeping the associated risk off their own balance sheets. Meta pioneered this approach for its data centers, disclosing in filings that such payments are "not probable" and therefore require no liability reserve. Broadcom extended the model to chip financing for companies like Anthropic, tying its exposure directly to whether customers continue paying.

What troubles bond traders is the scale and timing. With $70 billion in phantom liabilities floating in the market and AI chip debt expected to expand dramatically, the assumption that these risks are remote is increasingly being questioned. The proponents' argument — that chip demand will outstrip supply for years, that debt will be paid down over time, and that tech risk lands on balance sheets of companies with cash to absorb it — may not hold if the AI market faces a demand shock or a major customer defaults. The very fact that companies must explicitly state these risks are "not probable" suggests they are at least conceivable, and conceivable risks backed by tens of billions in hidden exposure can move quickly from theoretical to realized when market conditions shift.