Nvidia's Margin Pressure Is Micron's Revenue Boost

CFO Colette Kress spent most of Nvidia's (NASDAQ: NVDA) Q2 2027 earnings call on numbers that had gone up. Revenue rose 106% year over year. Data center sales jumped 117%. Next year's top-line growth is pegged at roughly 70%, and that is the scenario where supply runs out. In Kress's view, the market demand for artificial intelligence (AI) accelerators is rising faster than the manufacturing capacity.

Then she got to the gross margin line and walked everyone down a flight of stairs.

The second quarter showed a 75% gross margin, up from 72.5% in the year-ago quarter. In Q3, it should drop to 74%. Q4 will be a trough of 71% to 72%. The rebound in fiscal 2028 looks to be in the 72%-to-73% range.

Peak to trough, that is about 3.5 points. On a quarter guided to $108 billion, 1 point of gross margin is about $1.1 billion. These figures aren't rounding errors.

The explanation was refreshingly free of corporate spin. Memory prices blew past what Nvidia had modeled, they are still climbing, and rather than let investors stew on that for another quarter, Nvidia reset the numbers this week.

Follow the money, because it went somewhere

Gross margin does not evaporate. It relocates.

High bandwidth memory has grown to roughly 30% to 40% of an AI accelerator's build cost, up from under 20% two generations ago. Exactly three companies on earth make the stuff: Samsung (OTC: SSNLF), SK Hynix (NASDAQ: SKHY), and Micron (NASDAQ: MU).

Adding capacity takes time. Micron is adding manufacturing lines in Idaho, Virginia, and New York over the next eight years. Hynix plans to open four Korean memory facilities in 2029. Samsung accelerated the opening of its next factory from 2031 to 2029. The extra supply is coming, but not right away.

The math follows from there. Nvidia's cost of revenue is increasingly a payment to three global suppliers who cannot make enough of the product. The cash leaving Nvidia's gross margin line will land on someone else's income statement. That's great news for Micron, SK Hynix, and Samsung. All three stocks rose 3% to 5% after Nvidia's report.

Kress mentioned that Nvidia has deep relationships with all three memory makers and is working with them to expand capacity. Nvidia also signed a multiyear memory partnership with SK Hynix during the quarter. Those are the words and actions of a buyer securing a stable supply of components, not an aggressive price negotiator.

The detail worth catching

Nvidia is not eating sky-high memory costs forever. Kress said the fiscal 2028 recovery will happen "as executed price increases take effect in Q1" of that fiscal year. The increases are already negotiated. They simply do not reach the income statement for three more quarters.

So the honest read is a timing story, not a pricing-power question. Nvidia starts paying more now and charging more later. The upcoming margin dip is simply the space in between.

There is one more wrinkle. Nvidia shipped a small volume of earlier-generation Hopper 200 chips to China under U.S. licenses. These sales represented less than 1% of data center revenue, and Kress flagged those units as "dilutive to corporate gross margin." So the Chinese business isn't exactly zero, but it's also not a particularly lucrative revenue stream.

Image source: Getty Images.

The strange part is that this is bullish

Kress made an argument worth repeating, because it inverts the obvious reading. Memory scarcity is being driven in large part by the AI build-out itself. This is not a component that raises costs without an offsetting benefit; tighter memory supply is a symptom of the same demand surge driving Nvidia's growth.

The rest of the call backs that up. Nvidia guided fiscal 2028 to roughly 70% growth and called it supply-constrained. Customer forecasts point to demand roughly doubling. Nvidia's revenue opportunity per gigawatt of data center capacity has gone from about $18 billion in the Hopper era to $25 billion with Grace Blackwell to $40 billion with Vera Rubin.

A company that could sell twice as much as it can build has not lost its leverage. It has decided, for three quarters, to leave some of its pricing power on the table.

What to watch in November

Everyone is quoting the 74% margin dip today. The deeper Q4 trough is the more informative number. If the 71%-to-72% guide holds up when Nvidia reports on Nov. 17, it's a timing story with a tidy ending. On the other hand, a sliding range would mean that the memory makers are taking more profit than Nvidia budgeted for. In that case, the fiscal 2028 rebound turns into a loose forecast rather than a firm plan.

The other thing to watch is whether the price increases land. Nvidia has never struggled to charge more. It has also never tried it in a year when every major customer is quietly building its own accelerators.

In the meantime, the most profitable hardware business on the planet is temporarily routing a few billion dollars a quarter to the three companies that make the one component it cannot make for itself.

Anders Bylund has positions in Micron Technology and Nvidia. The Motley Fool has positions in and recommends Micron Technology and Nvidia. The Motley Fool has a disclosure policy.