Nvidia Just Committed $279 Billion to Its Suppliers. Read That Number Twice.

Nvidia reported its fiscal second quarter on August 26, and by every measure that fits in a headline it was a rout. Revenue of $96.2 billion, up 106 percent from a year ago. Data center revenue of $89.0 billion, up 117 percent. Guidance of $108 billion for the current quarter, which would be the first time the company has topped $100 billion in quarterly sales. Shares rallied the next session and pulled the broader tape up with them.

Then there is the part that did not make any headlines. Buried in the CFO commentary Nvidia filed alongside the results is a table of future commitments. One line in that table moved from $119 billion three months ago to $279 billion today. Nvidia’s explanation for the jump runs to a single clause. The increase was, in the company’s words, “primarily related to the procurement of memory.”

That is the real story of this quarter, and it is not entirely about Nvidia.

The Line Item That Should Have Led

Set $279 billion against the rest of the balance sheet and the scale becomes obvious. Nvidia held $56.6 billion in cash, cash equivalents, and marketable debt securities at quarter end. It issued $25.0 billion of senior unsecured notes during the quarter for general corporate purposes. Total future commitments across supply, cloud services, leases, equity investments, and capital expenditures came to $366 billion, with $120 billion of that falling in the remainder of this fiscal year alone.

Nvidia is not short of money. Free cash flow was $21.3 billion in the quarter and roughly $70 billion across the first half. But a company generating that kind of cash does not go out and lock in $279 billion of supply on a whim. It does so because it has concluded that the components in question will be scarce, expensive, or both, and that the risk of not having them dwarfs the risk of overcommitting.

Chief Financial Officer Colette Kress did not dance around it on the call. “We want to be direct about this, rather than let it linger as an open question,” she told analysts. “Memory scarcity today is being driven in large part by the AI buildout itself.” She described what she called extreme pricing conditions in memory and said the magnitude of the increases had run past the company’s own expectations.

You can watch the consequence work its way through the guidance. Gross margin came in at 75.0 percent for the quarter, both on a GAAP and a non-GAAP basis. The company guided the current quarter to 74.0 percent. Kress told analysts margins would keep sliding to a trough of roughly 71 to 72 percent in the fiscal fourth quarter before recovering into the low seventies next year, once price increases Nvidia has already negotiated finally reach the income statement. That is three quarters of margin compression at a business running the fattest margins in large cap hardware, and the cause is a commodity input.

Where the Margin Is Actually Going

If Nvidia is paying more, somebody is getting paid. Look at Micron.

Micron reported fiscal third quarter revenue of $41.46 billion for the period ended May 28, against $9.30 billion in the same quarter a year earlier. That is growth of roughly 346 percent. GAAP net income was $28.24 billion. Operating cash flow came in at $25.39 billion, up from $4.61 billion a year prior. Those are not the numbers of a cyclical commodity supplier having a good year. Those are the numbers of a business that has, at least for now, seized pricing power over the entire technology industry.

Memory has always been the most brutally cyclical corner of semiconductors, a business where capacity gets built at the top and prices collapse at the bottom with a reliability you could set a watch to. What has changed is the demand mix. High bandwidth memory used in AI accelerators eats far more wafer capacity per gigabyte than conventional DRAM, so every gigabyte diverted toward AI takes an outsized bite out of what is left for everything else. Reported capacity across the major suppliers is largely spoken for this year, and server memory contract prices have moved in increments that would have been unthinkable in an ordinary cycle.

For a couple of years the accepted framing of the AI trade was that Nvidia sat at the top of the value chain and everyone else was along for the ride. That framing is now incomplete. Value in this buildout migrates toward whoever holds the scarcest input, and at this moment the scarcest input is not logic. It is memory.

The Bill Does Not Stop at the Data Center

Here is the part that reaches people who have never bought a GPU in their lives. Nvidia’s own filing notes that edge computing revenue of $7.2 billion, up 27 percent from a year ago, was held back by “slower consumer PC sales that were tempered by elevated memory and systems prices.”

Read that again. The AI buildout is bidding memory away from consumer electronics with enough force that it turns up as a drag inside Nvidia’s own results. Every laptop, phone, console, and ordinary enterprise server that needs DRAM is now competing against hyperscalers who will pay nearly anything to get it. When people ask what AI capital spending costs the rest of the economy, this is one honest answer. It is not only capital diverted into concrete, steel, and transformers. It is a real price increase on a component sitting inside almost every electronic device sold.

And the spending is still climbing. Kress told analysts that capital expenditure among the top five hyperscalers is expected to rise to roughly $1.3 trillion next year from about $800 billion in 2026. Nvidia guided to approximately 70 percent revenue growth in fiscal 2028, well above the 44 percent analysts had been modeling, and framed that figure as supply constrained rather than demand constrained. Customer forecasts, she said, “point to our growth doubling next year.”

How to Think About This Without Getting Carried Away

None of this is a recommendation, and the correct posture toward numbers this large is curiosity mixed with skepticism rather than enthusiasm. A few frameworks are worth holding onto.

First, supply constrained guidance is a genuinely different animal from demand driven guidance. When a company says it could sell more if only it could build more, the variable that matters is no longer customer appetite. It is the supply chain, and supply chains break in ways that order books do not. Nvidia’s inventory rose to $31.6 billion from $25.8 billion sequentially, and days sales outstanding stretched to 60 from 45 on what the company described as extended payment terms with certain investment grade customers. Neither of those is an alarm. Both are worth watching.

Second, memory cycles end. They always have. The current pricing environment is pulling enormous capital into new capacity, and capacity arriving into a softer demand environment is precisely the mechanism by which memory booms have historically become memory busts. The open question is timing, and nobody knows the answer, including the people signing the checks for the new fabs.

Third, and most useful for anyone trying to read the buildout rather than trade it, that $279 billion line tells you something about how the companies inside this thing see the next few years. You do not sign a commitment of that size if you believe the whole edifice is about to deflate next quarter. You also do not sign it if you have any confidence that parts will be cheap and available whenever you want them. Both readings are true at the same time, and sitting with both is closer to an honest view than picking one.

The headline number was $96.2 billion. The number that told you more was $279 billion.

_______________________________________________________________________________________________________

This article is written for educational and informational purposes only and does not constitute financial or legal advice. The views and analytical frameworks presented draw on publicly available information and reported commentary from industry participants. Readers are encouraged to consult primary sources and form their own informed views on these complex topics.