Nvidia Doubled Its Revenue — So Why Isn't the Stock Moving?

On August 26, Nvidia reported the best quarter in its history: fiscal second-quarter revenue of $96.2 billion, up 106% from a year ago and an 18% sequential step, at a 75% gross margin. It beat its own forecast by roughly 6%, returned $26 billion to shareholders, and finished the period with $127 billion in trailing free cash flow and a net-cash balance sheet. Two weeks later the stock is down about 2% on the day and roughly flat over the past month — up 22% over six months but going nowhere even as the numbers get better.

That gap is the story. Nvidia is not struggling to prove the AI buildout is real. The market has simply stopped paying a premium for the proof.

The upside-surprise engine is running out

The machine that used to drive Nvidia's shares — the blowout beat — is winding down. Nvidia has topped management's own forecast for 14 straight quarters, but the size of those beats has narrowed steadily, from an average 22.8% above guidance in fiscal 2024 to 5.7% this quarter. Sell-side models have finally caught up with the ramp.

That compression is not fading demand. It is the market finally believing the growth. And beliefs, unlike surprises, only get repriced once. So the investor question shifts from "will Nvidia deliver?" — it will, at an extraordinary level — to "what happens to the forward numbers?" This quarter offers two concrete tells.

The first tell is a 3-point margin give-back

Gross margin was 75% in the quarter, essentially flat sequentially and up from a year ago. But management guided the next quarter down to 74%, expects it to bottom around 71%–72% in the quarter after that, and sees fiscal 2028 settling around 72%–73%, driven largely by rising memory costs. That is a disclosed, scheduled erosion of two to three gross-margin points.

Margins this high were partly a product of scarcity — a seller's market for an accelerator everyone had to have. As Nvidia ships more, whether through its new Vera Rubin platform or general supply growth, the scarcity premium normalizes. A 106% revenue print can hide the direction of the steady, smaller number underneath it.

The second tell is who is buying

Nvidia's growth is still overwhelmingly tied to a handful of hyperscale buyers — Microsoft, Alphabet, Amazon, and Meta together spent $166 billion on capital expenditures last quarter, up 87% year over year, and Jensen Huang calls the multi-year buildup a sustainable part of the cycle. But the more interesting signal is the non-hyperscaler slice of the data-center business, which management buckets separately: $40 billion in the quarter, up 25% sequentially and 138% year over year, fed by AI labs, enterprises, and sovereign customers. That is demand broadening beyond the megacap capex that everyone is already watching — the part of the cycle where a second or third customer base gets built.

It matters because the load-bearing architecture debate sits inside inference. Nvidia still dominates AI training, but inference is becoming a different, more contested market — the part of the cycle now growing fastest as trained models get served at scale, and where cost per token and latency, not CUDA lock-in, decide who wins. That is why Nvidia is now shipping dedicated inference silicon — its Groq LPX accelerator is in full production — and pushing software that it says cut token costs roughly fivefold on Blackwell. The moat that made Nvidia was built in training; inference re-opens the contest.

Nvidia does not need a monopoly to keep compounding if the market expands fast enough. But the shift to inference does erode the pricing power that made a 75% gross margin possible in the first place — which is the same operating force as the margin guidance above.

What has reached results, and what is still a claim

The revenue is real. The margin path is guided. The recurring-revenue answers — the new revenue-sharing model with "NeoClouds" that management says could bring in billions over the medium term — are announced claims, not reported results, and should stay out of the base case until they land on the income statement. The distinction is the whole discipline here.

So the long-term thesis is intact: roughly 70% revenue growth is guided for next fiscal year, the balance sheet is net cash, and the software layer that attaches to all that hardware is where the forward multiple will be decided. But an intact long-term thesis is not the same as an attractive near-term return. The market has already done most of the repricing — that is precisely why a record quarter trades flat. When a thesis has been fully digested, the discipline is to ask whether the remaining return curve still beats the rest of the AI trade, not to defend what has already been priced.

A stock that falls on a great report is usually an opportunity rather than a verdict. The judgment that matters now is which great report the market has yet to believe.