For four straight quarters, Nvidia has pulled off something that sounds impossible: it beat Wall Street’s revenue forecast every single time, and its stock fell every single time anyway. After the most recent one, in May, shares dropped 1.77% the next session despite a blowout quarter. That is the backdrop for Nvidia Q3 guidance, which the company released after the market closed on August 26 — and it explains why a record-shattering number came out to a shrug.
First, the raw numbers. Nvidia said it expects revenue of about $108 billion for the current quarter, give or take 2%. Analysts had penciled in roughly $104 billion. “Guidance” just means the company’s own forecast for the quarter ahead, and it usually moves the stock more than the results being reported, because investors are buying the future, not the past.
That $108 billion figure would be the first time Nvidia has ever cleared $100 billion in a single quarter. Fewer than ten companies in the S&P 500 have ever done it. And Nvidia got there while assuming zero data center revenue from China — a market it has effectively been shut out of by export rules. If China ever reopens, none of it is in the forecast.
The quarter just reported was strong too: $96.2 billion in revenue, more than double a year earlier, with the data center business alone bringing in $89 billion.
So where’s the catch?
Gross margin. That’s the share of each sales dollar left over after paying to make the thing — for Nvidia it has been running around 75%, which is extraordinary for a company that ships physical hardware. This quarter it held at 75.0%. Next quarter, the company says, it drops to 74%. And the quarter after that, it bottoms out somewhere between 71% and 72%.
The reason is memory chips. Nvidia’s newest systems need enormous amounts of them, and memory has gotten scarce and expensive. CFO Colette Kress was unusually blunt about the cause: the shortage is being driven largely by the AI buildout itself. In other words, the boom that is filling Nvidia’s order book is the same boom bidding up its input costs. I wrote about the other half of that squeeze recently — Nvidia has been passing some of it straight through to buyers in the form of higher server prices.
None of this suggests demand is cracking. Jensen Huang said Nvidia has secured enough supply for 70% revenue growth next fiscal year, and that actual demand runs well above that. Amazon Web Services separately committed to buying two million Nvidia GPUs.
But it does change what there is to argue about. For three years, the question was whether Nvidia could keep selling more. That question now has a boring answer: yes. The open question is what each sale is worth by the time it clears — and that matters more than usual, because Nvidia has been backstopping its own customers’ borrowing to keep those orders coming. Fat margins are what make that kind of guarantee affordable. Thinner ones make it a real cost.
My take: Nvidia still sells the shovels — it’s just started paying more for the steel.
Not advice. Just how I see it.