Nvidia's $530 Billion Isn't Hidden Debt. Read the 10-Q

A number has been circulating since Nvidia filed its latest quarterly report: $530 billion in "off balance sheet" commitments. The phrase "off balance" does not appear anywhere in the filing. The $530 billion does, sort of, if you add up three separate tables that measure three very different things. Reading them one at a time tells a more useful story than the total, and the most interesting risk in the document is not the biggest number.

Where the $530 billion comes from

Nvidia's Form 10-Q for the quarter ended July 26, 2026, the second quarter of its fiscal 2027, lists its obligations in three places in the notes to the financial statements.

The first table, "future commitments," totals $366 billion. The second, "additional commitments," totals $56 billion. The third summarizes the "maximum gross exposure" under guarantees: $108.5 billion. Add them and you get $530.5 billion. That is arithmetic, and it is correct. Treating the result as one pile of hidden debt is where the reading goes wrong.

Most of it is a shopping list

The largest single line, $279 billion, is supply and capacity: agreements with suppliers to secure components for data center systems, primarily memory and manufacturing capacity, over the next several years. That line rose from $119 billion the previous quarter. In three months, Nvidia more than doubled what it has lined up to buy.

The filing describes some of these agreements as cancelable, reschedulable or adjustable before firm orders are placed, with possible extra costs if they change. In plain terms, this is a manufacturer locking in parts it expects to need for products it expects to sell. It is a bet on demand, and a large one. It is not money borrowed.

The rest of the first table is smaller: $29 billion in cloud service agreements to support Nvidia's own research and open models, $25 billion in data center leases that have not started, $25 billion in committed equity investments in AI model makers and other private companies, and $8 billion in capital expenditures.

For scale, Nvidia reported revenue of $96.2 billion for that single quarter, up from $46.7 billion a year earlier, and net income of $59.7 billion. A company that size buying ahead is not, by itself, a warning sign.

The part worth reading twice

The guarantees are different. In August 2026, Nvidia agreed to provide credit support, capped at $105 billion in total, for a data center buildout by affiliates of SB Energy at the PORTS Technology Campus in Pike County, Ohio. The customer is an affiliate of OpenAI. The leases cover approximately 4.25 gigawatts of IT load across nine phases, the first expected in Nvidia's fiscal 2029, each on a 20 year lease.

The filing is specific about the mechanics. Each guarantee generally kicks in when its lease starts and grows as each phase is completed. Payments are triggered by certain tenant defaults. The exposure is expected to shrink over each lease as payments are made, and it is limited to defined portions of lease and power payments, not the full cost of the site. The guarantees end if OpenAI achieves a satisfactory credit rating. In exchange, the campus will host Nvidia AI infrastructure exclusively, with limited exceptions, and Nvidia holds an option to back about 3.8 more gigawatts as the site grows.

The supply line is a bet on demand. The guarantee is a bet on one customer.

The filing also explains why Nvidia is doing this at all. AI clouds and model makers, it says, have significant demand for computing but currently lack the ability to secure long term infrastructure contracts and investment grade financing. Nvidia is lending its balance sheet so its customers can build the sites that will run its chips.

Why the distinction matters

Lumping these together hides the one structural change worth watching. Buying memory ahead of demand is what chipmakers have always done, and if demand softens the damage shows up as write downs and cancellation costs. In the first half of fiscal 2027, Nvidia took a $0.4 billion charge on excess H200 inventory and purchase obligations after Chinese restrictions limited sales and demand for that chip faded. That risk is familiar.

A guarantee concentrated on one customer's data center is a newer kind of exposure for a hardware company. It ties part of Nvidia's financial outcome to the credit of a customer that is also one of the most important buyers of its products. The structure has protections, including caps, phase timing and a release clause tied to a credit rating, and none of it starts until fiscal 2029. It still means the company selling the shovels is now cosigning for one of the miners.

The viral number made the story sound like hidden debt. The actual filing describes something more interesting: a supplier so central to the AI buildout that it has started financing its own demand. Whether that looks prudent or reckless will depend less on the size of the tables than on whether the customers behind them keep growing into the capacity being built for them.