Nvidia's earnings: the five numbers that actually matter
The headline beat is the boring part. The story is in data-center growth, gross margin, China, and one quiet line in the guidance.
Nvidia reported. The stock did the usual jittery dance after-hours — up, down, up again — and the financial press did the usual ritual of yelling at the top-line beat. We're going to do something less fun and more useful. Five numbers, what they say, what they don't, and what we'd be watching next quarter.
The plain-English bit
Nvidia makes the chips that train and run modern AI models. It also sells a software stack (CUDA) and increasingly the whole computer that goes around the chip. The reason a single chipmaker has been worth more than the entire German stock market at times is that essentially every AI company on earth is buying from them, more or less at the price they ask.
1. Data-center revenue: $26.4B, up 152% YoY
This is the only number that really matters quarter-to-quarter. Gaming is a footnote now; data center is the company. The 152% year-over-year growth rate is, frankly, ridiculous for a business of this size. It also obscures the sequential growth rate — quarter-over-quarter, the segment grew about 16%. Still excellent. Just not exponential.
The thing to watch: sequential, not annual. Annual comparisons get easier — or, more cynically, more flattering — for another two quarters. After that, the comp gets brutal. If sequential growth keeps decelerating from the 30s to the high teens to the single digits, that's a story.
2. Gross margin: 75.1%
Software companies have margins like this. Chip companies don't. The fact that Nvidia is running at three-quarters gross margin while shipping physical hardware tells you it's pricing closer to "what the market will bear" than "cost plus a reasonable markup." That's great for shareholders today and a flashing yellow light for shareholders in three years.
Why? Because 75% gross margins are an open invitation to competition. AMD is shipping. Custom silicon from the hyperscalers — Google's TPU, Amazon's Trainium, Microsoft's Maia — is real and getting better. The question is not whether Nvidia loses share. It's how fast, and at what margin.
The receipts
Last cycle's high-margin semiconductor stars — Intel in the early 2000s, Qualcomm in the 2010s — both eventually surrendered 10-20 percentage points of gross margin as customers built their own chips or as competition caught up. Nvidia is not those companies, but it is selling into the same physics.
3. China revenue: 12% of total, down from 25%
US export controls now restrict Nvidia's top-tier chips from going to China, and the company has been selling a deliberately dumbed-down product into that market. China revenue is roughly half what it would have been on the old run-rate. What's surprising is how little it has mattered to the headline numbers, because the rest of the world has more than absorbed the slack.
The risk is asymmetric: the export rules can tighten further, but they're not loosening. Assume the China line stays a low-and-shrinking percentage of the total. The bull case has to work without it.
4. Operating expenses: up 38% YoY
The most underrated line on the page. Nvidia is hiring aggressively, expanding its software org, and pushing into networking (Mellanox), inference (the new B-series rollouts), and enterprise software (NIM, NeMo, et al.). This is exactly what you'd want a company in their position to do — re-invest the gusher of cash into widening the moat. But it also means operating leverage is no longer the easy story it was 18 months ago.
5. The one line in the guidance that nobody asked about
Buried in the call: "We expect networking to be the fastest-growing portion of our data-center business in the back half." This is the quiet tell. Selling chips is one thing; selling the whole rack — switches, optics, software, the box you plug into the wall — is a much bigger TAM and a much stickier customer relationship. If networking is genuinely accelerating into 2H, the narrative shifts from "Nvidia, the AI chip company" to "Nvidia, the AI systems company." Those are valued very differently.
The boring lines on an earnings page tell you what the company will be in three years. The headline lines tell you what it was three months ago.
So what's the take?
This was, on the merits, a great quarter. The story is intact: AI capex is still being spent, the company is still printing, the margin is still absurd. But every great chip cycle in history has had two phases: the one where the supplier has all the leverage, and the one where the customers figure out alternatives. Nvidia is firmly in phase one. The market is paying for phase one to last forever. It won't, but it might last longer than the bears expect — long enough that the bears get tired of being right early.
We are not going to tell you whether to buy the stock. We will tell you that if you own it, the things to watch next quarter are sequential data-center growth, gross margin direction, and whether networking is named in the prepared remarks before the analysts have to ask. If yes to all three, the run continues. If two of three slip — pay attention.
Disclosure: One member of the Money Market desk holds shares of NVDA, purchased before this site was launched. Specific figures in this piece are illustrative for a hypothetical recent quarter and do not represent actual reported results. For real figures, see Nvidia's most recent 10-Q on SEC EDGAR.