Nvidia's Q2 Was a Blowout. The Red Flag Is Margin — and It's Not About Demand

On August 26, Nvidia (NVDA) gave the market exactly what it wanted. Revenue of $96.2 billion beat the roughly $92 billion consensus, up 106% from a year ago. Adjusted earnings of $2.22 a share came in about 6% ahead of the Street. Data Center hit $89 billion, up 117%. Management then guided the October quarter to $108 billion — another record — while assuming zero China compute. The stock climbed, the "Strong Buy" chorus stayed put.

So why would anyone call this quarter a red flag? Because buried in the same release was the first break in the margin story that has done as much for this stock as the revenue line.

Nvidia guided gross margin down to 74% for Q3, called a bottom of 71%–72% in fiscal Q4, and told analysts to model 72%–73% for next year. That last number is the quiet part: the Street had been penciling in 75.1% for fiscal 2028.The company delivered 75% in the fiscal fourth quarter and 75% again this quarter — the mid-70s line was the ceiling the bull case stood on. It is now a floor being walked down, and management drew the path itself.

The important thing is why it's happening, because that sets whether this is a warning or a footnote.

The money is moving to someone else — and it isn't Nvidia's choice

This is not a demand problem. Demand is the least of it. The margin is being squeezed by the one input every accelerator in this cycle is bottlenecked on: high-bandwidth memory. Nvidia's own suppliers — SK Hynix and the other memory makers — are in a supply crunch of their own, and when the scarce component holds the leverage, the scarce component captures the margin. Guidance was cut on higher memory costs and the Rubin ramp, and Nvidia has been raising prices on its own hardware to claw some of it back.

You can see the transfer in the order book. Total supply commitments have swelled to $279 billion, and they are mostly memory commitments for the Vera Rubin platform. Read that figure the way it should be read, as a dual signal. On one side it is extraordinary conviction — Rubin is already in production at Google Cloud, Azure, and Oracle, and Nvidia calls it the fastest ramp in its history, backed by inventory that has climbed to about $32 billion. On the other side, $279 billion of locked-in memory means Nvidia has already accepted the input cost that decides its margin for the next several quarters, before a single Rubin system is recognized.

Put a number on it. At the $108 billion quarterly run rate management just guided, roughly three points of gross margin is about $3 billion a quarter — on the order of $13 billion a year at today's scale — that now flows to memory suppliers instead of Nvidia's operating line.

The leverage that shows up in cash, not just margin

The margin guidance is the headline red flag, but it is the visible tip of the same decision. Nvidia is running a heavier balance sheet to keep this ramp financed. Free cash flow stepped down to $21.3 billion from $48.6 billion the quarter before, on longer payment terms for large deals as customers stretch out big purchases. Days sales outstanding widened to 60 days.The company issued $25 billion in bonds in the quarter and committed $105 billion in guarantees on an Ohio facility leased to OpenAI.

None of this is a solvency concern for a company generating this much cash. But it is a shift in who bears risk across the AI buildout, and it compounds the margin story rather than offsetting it. The bull case has always been built on Nvidia collecting a huge share of the value created by AI. The message of this quarter is that a meaningful slice of that value is being routed to the memory oligopoly and into the financing structures that hold up the data-center buildout, not into the dividend and buyback line investors have been trained to watch.

What this means for the stock, and what it doesn't

This is the case for taking the red flag seriously without turning bearish — which is the right posture. The demand story for Nvidia is intact; management still guides roughly 70% growth next year with supply constrained through fiscal 2028. A company compounding revenue at that rate can absorb a three-point margin haircut and still grow absolute profit, because volume covers the per-dollar give-up. The question the margin guidance raises was never whether Nvidia stays important.

The question is the return curve. The market has already navigated to this being a ~16x forward-fiscal-2028 earnings stock, per analyst commentary, on a market cap north of $5 trillion. At that price the near-term path is no longer driven by whether demand holds — it's driven by whether growth outruns a margin profile that management itself has just said settles in the low 70s. A three-point margin step-down, layered onto an already-rich multiple, flattens the near-term earnings-per-share curve even when the top line keeps compounding.

That is the honest read of this quarter. Margin compression here is not a demand problem and not a Nvidia-execution problem — it is the cost of the fastest product ramp in the company's history, paid to the suppliers who hold the scarce input. If memory costs stay elevated past the Rubin ramp and revenue growth normalizes toward 30%–40%, that margin reset becomes the ceiling on the stock's next move. If Rubin ramps on schedule and demand keeps pulling supply, volume offsets the margin give-up and today's reset is the clearing event that lets the compounding resume.

The discipline is not to sell on the beat-and-raise, and not to ignore the guidance. It's to know which variable this quarter actually changed. It changed the near-term return curve, not the long-term thesis — and on a multiple like this, that distinction is the whole investment case.