Nvidia's supply commitments hit $279B. Read who carries the risk.

Nvidia reported $96.2 billion of revenue for the quarter that ended 26 July 2026, up 106% from a year earlier, and guided the current quarter to $108 billion. That is the number the headlines carried. The same day, 26 August 2026, the company published its CFO commentary, and the back half of that document tells a different story than the press release.

In three months, Nvidia's supply and capacity commitments went from $119 billion to $279 billion, "primarily related to the procurement of memory." Customers now take 60 days to pay, up from 45 the quarter before. Operating cash flow fell to $24.1 billion from $50.3 billion, in a quarter when revenue rose 18% sequentially. Nvidia issued $25 billion of senior unsecured notes while holding $56.6 billion of cash and marketable securities. And it capped a guarantee at $105 billion behind one customer's twenty-year data center leases.

Put those five lines side by side and the quarter reads differently. Demand is real. Nvidia is also funding a growing share of it, paying suppliers earlier, collecting from customers later, and standing behind the buildings that will house the chips.

$92 billion of the $279 billion lands within six months

Nvidia breaks its future commitments out by fiscal year. Its fiscal 2027 ends in late January, so "remainder of 2027" covers the two quarters between the filing and then.

Two quarters, $92 billion of purchase obligations, against $108 billion of guided revenue for the first of them. Then $87 billion in fiscal 2028 and $88 billion in fiscal 2029, after which the schedule drops to $6 billion and $5 billion. Nvidia has locked down roughly three years of inputs and left the fourth open.

Supply is one row of a larger table. Add cloud service agreements at $29 billion, data center leases not yet commenced at $25 billion, equity investments at $25 billion and capital expenditure at $8 billion, and total future commitments reach $366 billion. A second table adds $36 billion of AI cloud agreements and $20 billion of leases Nvidia expects to hand off to third parties. Guarantees sit outside both tables, at a maximum gross exposure of $108.5 billion.

Set against a company that booked $63.7 billion of operating income in a single quarter, none of that is reckless. It is a long way from the asset-light chip designer people still picture.

The money went into memory because memory got expensive

Nvidia named the reason itself: procurement of memory, secured with what it calls "our extensive network of suppliers to secure the critical components needed to meet demand for the next several years."

Here is what those suppliers have been charging. TrendForce put the increase in conventional DRAM contract prices at 93% to 98% quarter on quarter in the first quarter of 2026, with industry revenue up 81% to $97 billion over the same three months, and forecast a further 58% to 63% rise in the second quarter. By its 3 July 2026 update the rate of increase was cooling, to a projected 13% to 18% for the third quarter, with NAND flash up 10% to 15%.

Cooling from a doubling is still an increase. And the squeeze has a cause you can point at. High-bandwidth memory, the kind that sits next to an AI accelerator, ate about 18% of total DRAM wafer input at the end of 2025. TrendForce expects roughly 22% by the end of 2026 and about 30% by the end of 2027. Every wafer that goes to HBM is a wafer that doesn't go to the DDR5 in a laptop.

Which shows up in Nvidia's own results. Edge Computing revenue grew 27% year on year to $7.2 billion, and the company attributes the drag to "slower consumer PC sales that were tempered by elevated memory and systems prices." The same shortage that pushed Nvidia to commit $279 billion is making consumer machines dearer, and Nvidia is reporting it as a headwind in its own segment.

Gross margin tells the other half. Nvidia held 75.0% last quarter and guided the current one to 74.0%, plus or minus 50 basis points. One point of margin on $108 billion of revenue is about $1.1 billion. That is the memory bill arriving.

Fifteen extra days of customer credit

Accounts receivable finished the quarter at $63.1 billion, with days sales outstanding at 60. The quarter before, it was 45. Nvidia's explanation, in full: "due to extended payment terms on large, multi-quarter agreements with certain investment-grade customers."

Fifteen days on this revenue base is real money. At roughly $96 billion a quarter, each day of DSO is about $1 billion of cash sitting on someone else's balance sheet instead of Nvidia's.

The cash flow statement shows the effect. Operating cash flow came in at $24.1 billion, up from $15.4 billion a year ago, and down from $50.3 billion in the previous quarter. Revenue rose 18% sequentially over that same stretch. Nvidia points to working capital and cash taxes. Inventory is part of it too, at $31.6 billion against $25.8 billion three months earlier, as the company builds ahead of the Vera Rubin launch.

Then the financing line. Nvidia issued $25 billion of senior unsecured notes during the quarter "to be used for general corporate purposes," and returned nearly $26 billion to shareholders through buybacks and dividends. A company with 75% gross margins and $56.6 billion of liquid assets borrowed $25 billion. You can read that as cheap capital taken opportunistically. You can also read it next to the $92 billion of purchase obligations due before February.

The $105 billion cap

The largest single number in the filing is contingent. In August 2026 Nvidia entered guarantees providing credit support on land, power and shell for about 4.25 gigawatts at SB Energy's PORTS-Pike campus in Pike County, Ohio. The site will host Nvidia infrastructure exclusively, under twenty-year leases to OpenAI. Nvidia's guarantee obligations are "capped at a total of $105 billion," take effect in phases as data centers come into service, with the first expected in fiscal 2029, and shrink as OpenAI makes its lease payments. Nvidia also holds an option to extend the same support across another 3.8 gigawatts.

Precision matters here, because a lot of the coverage of the 17 August announcement called it financing. Nvidia is not writing a $105 billion cheque. It is capping what it would owe if OpenAI stopped paying rent on a twenty-year lease. The exposure is contingent, phased, and declining. It is also the amount Nvidia was willing to put behind one customer's ability to pay, It also dwarfs the $3.5 billion of similar guarantees covering Nvidia's other AI cloud partners.

Nvidia attaches its own projection to the site: each generation of infrastructure deployed there could mean roughly 1.5 million GPUs and $150 billion to $200 billion of Nvidia revenue. That is a company forecast, not booked business, and it depends on a campus whose first phase is still more than two years out.

Nvidia has already written the diagnosis

The most useful sentence in the document is Nvidia's own account of why any of this is necessary:

"AI clouds and model makers are seeing extraordinary demand for AI infrastructure, yet many are growing faster than their balance sheets and long-term credit profiles can support."

That is the vendor explaining that its buyers cannot fund their purchases. Everything else in the commentary follows from it: the longer payment terms, the lease guarantees, the arrangements to help customers "secure the land, power and data center capacity needed to support their growth," the equity stakes in model makers and infrastructure financiers.

Sixteen days before the print, Nvidia announced platforms with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR to mobilize more than $500 billion of third-party capital for AI infrastructure. The structure is designed to keep the buildout off Nvidia's own books. Those agreements are memorandums of understanding: no partner has disclosed a dollar figure, and no first project has been named. In the quarter that closed before the announcement, Nvidia's balance sheet absorbed $160 billion of additional supply commitments, $25 billion of new debt, $108.5 billion of guarantee exposure and fifteen more days of customer credit. The off-balance-sheet plan is real and mostly still on paper. The on-balance-sheet commitments already cleared.

Anyone tracking how AI infrastructure gets paid for has seen this shape before, in the compute rent booked years ahead of the hardware by companies with their own funding gaps.

What the filing does not show

It does not show distress. Gross margin is 75.0%. Operating income was $63.7 billion in the quarter. Net gains from equity securities added $7.8 billion. The company returned nearly $26 billion to shareholders and still ended with more cash than it started.

Purchase commitments are contracts to buy inputs Nvidia expects to resell at 74% to 75% margins, and they are not debt. The receivables went to counterparties Nvidia describes as investment grade. The guarantees are capped, contingent, phased and amortising. Read charitably, the whole package is a company using a very strong balance sheet to clear bottlenecks that would otherwise cap its growth, and locking in memory before prices rise again.

The honest version is that both readings are true at once, and they differ only in what happens if demand slows. If it holds, Nvidia bought three years of supply ahead of a price rise and financed customers who were always good for it. If it slows, the $279 billion of purchase obligations, the $31.6 billion of inventory, the $63.1 billion of receivables and the $105 billion cap all sit on Nvidia, and by extension on the index funds that carry it. That risk did not disappear. It moved.

What to watch next quarter

• Gross margin against the 74.0% guide. It is the cleanest read on whether Nvidia is absorbing memory costs or passing them through.
• DSO. If 60 days becomes 70, extended terms are a policy rather than a one-off.
• The supply schedule. Whether the $92 billion due by January converts into revenue or into inventory.
• A named project from the $500 billion platforms. Until one exists with a number attached, the buildout is being financed on Nvidia's books.
• DRAM contract prices for the fourth quarter. The whole memory commitment is a bet on the direction of that line.

And keep the distinction between a quarter's revenue and the terms that produced it. It is the same discipline that applies when a private AI company reports a revenue milestone and everything depends on how it is counted.

Sources

• NVIDIA CFO Commentary on Second Quarter Fiscal 2027 Results, 26 August 2026
• NVIDIA Announces Financial Results for Second Quarter Fiscal 2027, 26 August 2026
• TrendForce: Rapid Contract Price Surge Drives 1Q26 DRAM Industry Up 81% QoQ, 1 June 2026
• TrendForce: Tight DRAM Supply Gives Suppliers Greater Pricing Power in HBM, 2 June 2026
• TrendForce: AI Server Demand Continues to Support Memory Prices in 3Q26, 3 July 2026
• NVIDIA Partners With Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR, 10 August 2026
• CNBC: Nvidia backing $105 billion in financing for OpenAI data center in Ohio, 17 August 2026

This is not financial advice.