Nvidia's most recent 10-Q contains a figure that has drawn more attention than the revenue line: roughly $530 billion in off-balance-sheet financial guarantees and commitments, up from about $184 billion a quarter earlier - an increase of more than 180% in a single quarter. The total spans six categories, including supply commitments, data-center take-or-pay agreements and cloud service leases. Analysis of the filing by the research firm SemiAnalysis, first circulated through Chinese financial outlets including Wallstreetcn and Gelonghui, has quickly become the reference point for investors trying to size Nvidia's real exposure.
Three things drove the increase. Supply commitments rose sharply, mostly on memory procurement. Data-center backstop agreements expanded, including support for SB Energy's PORTS-Pike campus in Ohio, whose capacity is earmarked for OpenAI. And two brand-new line items appeared on the disclosure list for the first time: $36 billion of AI cloud services agreements - effectively backstop commitments to neocloud operators - and $20 billion of data-center leases that Nvidia signs up front and later transfers to neocloud counterparties.
The structure matters more than the size. Nvidia is no longer only a chip supplier; it is becoming a credit backstop for the AI buildout, underwriting the financing chain that turns other people's data centers into running capacity. That role extends Nvidia's financial commitments far beyond what a conventional balance-sheet reading captures, and it embeds the company in the solvency of customers whose business models are still being proven.
SemiAnalysis argues Nvidia has the earnings base to carry it. The firm expects Nvidia's EBITDA to reach about $450 billion in the fiscal year ending January 2028, and estimates the company could expand off-balance-sheet obligations beyond $1 trillion in the coming years while maintaining a margin of safety. The firm is explicit that this is a framework for the upper limit of what Nvidia could absorb, not a forecast.
A quieter disclosure compounds the question. In late September, when Nvidia raised its buyback authorization to $235 billion, its language changed to returning cash "in the form of free cash flow net of strategic uses" - effectively folding strategic equity investments into the deduction. Under that narrower reading, adjusted free cash flow for the first half of the fiscal year shrinks from the officially reported $69.9 billion to roughly $21.7 billion, a decline of nearly 70%, according to a Wall Street Journal analysis cited in the coverage. Nvidia also added $24.9 billion of long-term debt during the period to help fund $31 billion of incremental buybacks.
The circularity is hard to miss. Nvidia holds stakes in 13 publicly traded and 229 private companies spanning data-center construction, model development, energy supply and hardware manufacturing - including OpenAI and Anthropic. Many of those investees are simultaneously chip customers: Nvidia injects capital, and the capital comes back as GPU orders. That loop has been debated for a year; $530 billion of off-balance-sheet exposure gives it a balance-sheet dimension that can no longer be treated as a footnote.
The risk case is straightforward. If AI infrastructure investment slows, supply commitments can generate cancellation or adjustment costs, and data-center leases, cloud agreements and guarantees become contingent on customer performance, replacement tenants and asset recovery. Some supply agreements are cancellable, deferrable or adjustable, which softens the hardest landing - but only to a point. SemiAnalysis frames the question without pretending to answer it: can Nvidia honor every promise, on and off the balance sheet, while keeping a margin of safety if the cycle turns?
Comments (0)
Log in to join the discussion
Log InNo comments yet