AI

Nvidia guided to $108 billion. The number that matters is $279 billion.

Revenue hit $96.22 billion, data center grew 117%, and management told investors to expect $108 billion next quarter with zero China compute in the figure. Buried in the same filing: supply commitments have swelled to $279 billion, mostly memory for Vera Rubin.

N Noah · The Sharp Brief · August 26, 2026 · 3 min read
Deep aisle of a dark warehouse stacked with unmarked crates and server hardware

Nvidia closed its fiscal second quarter with $96.22 billion in revenue, up 106% from the $46.7 billion it booked a year ago, and adjusted earnings of $2.22 a share against the $2.09 Wall Street expected. Data center revenue alone reached $89.02 billion, a 117% jump. Data center networking — the switching and interconnect that stitches racks into clusters — grew 138%.

Then management guided the current quarter to $108 billion, plus or minus 2%, with non-GAAP gross margins near 74%. Consensus sat around $104 billion. The guide assumes no data center compute revenue from China whatsoever; Hopper shipments there came to less than 1% of data center revenue last quarter, so anything licensed lands on top of the number rather than inside it.

Every headline figure cleared. The stock slipped in extended trading anyway. That is now five consecutive quarters of beating estimates, and four of the previous five reports were followed by a decline. When roughly 70% growth is already priced in, a beat is the baseline, not the news.

The line nobody read

Further down the filing sits a figure that drew a fraction of the attention: supply commitments have grown to $279 billion, largely tied to memory procurement for the Vera Rubin platform.

Supply commitments are not customer orders. They are contractual obligations running the other direction — what Nvidia has agreed to buy from its own suppliers. Fab time, high-bandwidth memory, advanced packaging capacity, all reserved well ahead of the revenue those components will eventually produce. It is the clearest financial statement Nvidia has made about what it thinks the next two years look like, and it is far more expensive to reverse than a sentence on an earnings call.

Memory is the reason the number is that large. HBM supply has been the binding constraint on AI server production for over a year, and it is already pushing server prices up roughly 15%. Locking in memory years out is how you avoid being rationed. It also means Nvidia now carries a quarter-trillion dollars of committed spend against demand that has to keep showing up.

Our take: The $108 billion guide is a forecast. The $279 billion is a wager, already placed. Guidance can be walked back next quarter at the cost of a bad afternoon; supply commitments are signed contracts with suppliers who have their own capacity plans. Read the two together and the message is that Nvidia believes the constraint is supply, not demand — and it has spent accordingly. That is a genuinely bullish signal about the buildout and a genuinely concentrated risk if the buildout ever pauses. Both things are true at once, and only one of them shows up in the headline.

What changed underneath

Chief executive Jensen Huang framed the quarter on the call as a shift in what customers are actually buying: “AI has reached its inflection point. It’s doing useful work. Its tokens are productive and profitable. Now, compute is revenue.” He also described a broader customer base than a year ago, when a single lab was driving the buildout — now multiple frontier labs, an open-model ecosystem and physical AI are scaling in parallel.

That matters more than it sounds. Concentration was the standing bear case: too much of the revenue traceable to too few buyers. Vera Rubin reaching full production at Google Cloud, Microsoft Azure and Oracle Cloud Infrastructure spreads the base without shrinking the ticket size. It also explains why the memory commitments are front-loaded — the next platform is already shipping into that base.

What to watch

Nvidia is a roughly $5.1 trillion company that just told the market it expects to sell $108 billion of hardware in three months. The more revealing disclosure is that it has already promised to buy $279 billion of parts to do it.

Advertisement

Get the day, decoded — at 7 PM ET

The Sharp Brief: AI, money, business & performance in five sharp minutes. Free.

Free bonus: subscribe today and The 2026 AI Playbook (PDF) lands with your welcome email.

Recommended by 5+ newsletters across AI, markets & business.