Nvidia no longer just sells chips: it increasingly backs the people who buy them with its own balance sheet. Its latest 10-Q (quarter ended July 26, 2026) lists commitments and guarantees that analysts add up to about US$530 billion, and SemiAnalysis warns they could exceed US$1 trillion in the coming years.
What the 10-Q Says, and What It Doesn't
A caveat up front: the US$530 billion figure is a sum computed by third parties, not a total Nvidia declares. According to FourWeekMBA's analysis, the 10-Q presents three tables that add up to US$530.5 billion; the filing doesn't use the term "off-balance-sheet" and doesn't state a single total. The components reported by secondary sources:
These figures come from secondary coverage of the 10-Q; if you plan to cite them in a formal document, check them against the filing on SEC EDGAR.
What Drove the Jump in One Quarter
According to Deep News (citing the 10-Q and SemiAnalysis), the increase came mostly from larger supply commitments, chiefly to buy memory; from backstop agreements for data centers, including SB Energy's PORTS-Pike campus in Ohio, whose end customer is OpenAI; and from two new categories: a US$36 billion AI cloud services agreement backing "neocloud" operators, and a US$20 billion data center lease that Nvidia signed first and then transferred to a neocloud counterparty. The guarantees for the Ohio campus are reported with a cap of US$105 billion, triggered in phases as conditions are met; a separate US$250 billion figure for the OpenAI project is described only as a pending negotiation, not a number from the filing.
SemiAnalysis's Question
SemiAnalysis projects these obligations could exceed US$1 trillion in the coming years, clarifying that this is not an earnings forecast but a framework for a theoretical ceiling. Its underlying question: could Nvidia meet all of its on- and off-balance-sheet commitments during an AI infrastructure slowdown while keeping an adequate margin of safety? The source says there is no definitive answer. Investor Michael Burry, for his part, observed that Nvidia would essentially be financing OpenAI's purchases of Nvidia hardware.
For operational context, on the same day Nvidia announced a capacity expansion with six suppliers (per GlobeNewswire), while Business Insider reported, via Benzinga, that some server racks bought by Amazon are sitting in warehouses waiting on electricity at its data centers: the bottleneck is no longer just the chip.
What This Means for a Company Buying AI
- Ask your cloud GPU provider what happens to your contracted capacity if its financing tightens.
- Avoid long-term compute commitments without exit clauses or portability between providers.
- Design workloads so they can run on more than one provider and model (inference abstraction).
- Follow these filings as a cycle indicator: supply commitments are a thermometer of expected demand.
Questions about designing an AI architecture that's portable across providers: info@guatemalia.com.