Here is the contradiction most people missed this week.
Nvidia beat. It always beats. It has now beaten its own guide for thirteen straight quarters. And per netzender.com, for the last four of those beats, the stock reportedly fell the next day anyway.
So if you were watching the $96 billion print land — reportedly up 106% year over year, data center up 117% — and waiting for that to move the stock, you were watching the wrong number entirely.
Last week we said exactly this: the quarter was pre-sold. The Street had settled around $91.9 billion against Nvidia's own $91.0 billion guide, a consensus barely 0.9% above the midpoint. The print was priced in weeks before it printed.
We said only two things could move the stock: the guide and the margin. This week resolves that open thread — the one about whether the August 26 report would finally break the post-earnings decline pattern.
It broke. And it broke on a single sentence about next year.
Colette Kress reportedly said Nvidia expects revenue to grow 70% in fiscal 2028 — the year that runs February 2027 through January 2028. Against a fiscal 2027 reportedly tracking toward roughly $425 billion, that math points at approximately $700 billion. Knocking on the door of a trillion.
That is the event. Not the $96 billion. The $96 billion was the non-event we told you it would be — a thirteenth consecutive beat that, on its own, nobody rewards anymore.
Now look at what that 70% number actually exposes, because this is the part that reframes the whole story.
A year ago, the average sell-side consensus for Nvidia's fiscal 2028 revenue was about $310 billion. The guide implies roughly $700 billion. The Street was wrong by around $400 billion — on a single company, on a one-year view, twelve months out.
And it wasn't as if analysts stood still. They spent the entire year revising upward. Even walking into this call, consensus sat around $574 billion. Still short of the guide by roughly $125 billion after a year of nothing but upward revisions.
That is the real story of the curse breaking. It didn't break because demand showed up — demand was never the question. It broke because the guide, for once, was the event, and the guide detonated the sell-side model that had been chasing this curve and missing it for four straight quarters.
The curse didn't break because demand showed up. It broke because the guide, for once, was the event — detonating a sell-side model wrong by $400 billion.
Then Huang went further. He said demand "is much greater than 70%" — that the number is a supply ceiling, not a demand ceiling. Read plainly: 70% growth is the floor, capped by how much TSMC and the memory makers can physically ship, not by how many buyers exist.
The tell inside that was quieter but bigger. Huang noted that Nvidia has "never forecasted" a year in advance, but now has "a lot greater visibility" across the supply chain to do so.
That is the actual regime change. A company that historically refused to guide a year out just did — which is why the market treated the number as information rather than aspiration.
The reaction was violent by Nvidia's standards. Shares jumped roughly 9%, reportedly adding about $440 billion in market cap in a day. After four straight post-earnings selloffs on quarters it met or beat, that is the pattern snapping.
The neocloud complex moved with it. Nebius rallied, and per netzender.com the chip stocks Broadcom and Intel rose, while CoreWeave traded flat. The read-through was immediate: if Nvidia can ship it, everyone who resells Nvidia capacity is in the money.
Here is where the story gets uncomfortable, and where the +9% pop papers over the thing that actually matters.
Even after the move, Nvidia trades at about 25 times forward earnings. On the guide, that is almost absurdly cheap — one bull's back-of-envelope reportedly put it at roughly 11 times fiscal 2028 earnings if a 64% operating margin holds against $700 billion.
Now the number that should stop you. Through all of this — through a year of making a mockery of analyst estimates — the stock has barely outperformed the S&P 500 and has actually underperformed both the Nasdaq-100 and the semiconductor ETF.
Sit with that. The single company most responsible for the AI trade, printing 106% growth and guiding to 70% more, has lagged the index it dominates and the sector it defines.
That gap — between operating performance and stock performance — is the tell. It means the buy-side is quietly carrying a materially different model than the sell-side.
The sell-side model shows Nvidia compounding revenue at mid-to-high-60s operating margins. The multiple says the buy-side doesn't fully believe it will hold, or won't underwrite it at a rerate. The curse broke, but the wall of worry it was climbing didn't move.
So what is the worry, exactly? Not demand. It's the shape of the demand.
Nvidia's own mix hints at what it's fighting. Hyperscaler revenue doubled year over year, but the AI Clouds, industrial and enterprise bucket reportedly grew about 138% to $40.3 billion — Nvidia visibly trying to widen its base beyond the four cloud giants.
It's leaning hard on neoclouds to do it. Nvidia now reportedly expects neoclouds to exit 2026 with around 8 gigawatts of installed capacity, up from roughly 3 the year before — an enormous incremental pull, given AWS, the largest hyperscaler on earth, reportedly ran an estimated ~15 GW in 2025.
But here is the question the buy-side keeps asking, and it's the reason the multiple won't budge: who is actually in the neoclouds' backlog? The same hyperscalers.
If the hyperscalers pull their own capacity closer to their own demand curve over time, the neocloud demand that's inflating Nvidia's fiscal 2028 could partially evaporate — because it was hyperscaler demand wearing a neocloud costume.
The concentration cuts the other way too. Much of the frontier-lab demand traces back to a handful of names — and those names are deliberately building heterogeneous fleets.
Nvidia reportedly flagged that OpenAI's commitments represent 12 GW of Nvidia compute. But OpenAI has also reportedly signed 6 GW with AMD, 10 GW with Broadcom, and 2 GW with Amazon — leaving Nvidia's share of that footprint at roughly 40%. Anthropic's mix is reportedly worse for Nvidia, with its share closer to 20%.
The frontloading is real — the early gigawatts skew Nvidia — but the stated intent of the biggest buyers is to migrate share away over the medium term. A curve concentrated in a few labs actively courting alternatives is a curve the buy-side will not pay a premium to own, however steep it looks today.
That is the resolution to the whole apparent paradox. The reason Nvidia trades at 25x while guiding to 70% growth is not that anyone doubts the 70%. It's that they doubt the durability and diversification of what's underneath it.
Which is why the curse breaking matters less than it looks. The four-quarter selloff pattern was a symptom — the market repeatedly refusing to reward beats it had already priced. The guide finally gave it something unpriced, and it responded.
But a $440 billion day doesn't rerate a stock that's spent a year underperforming its own sector. It just closes a small gap between an impossible print and a skeptical multiple.
The demand question is answered — emphatically, by Huang's own "much greater than 70%." The question that's left is the one a guide can't settle: whether the buy-side will ever underwrite a curve this concentrated, from a sell-side that's been wrong on it by $400 billion, at anything more than the wall-of-worry multiple it's paying right now.
For now, they'll take the beat, take the pop, and keep the discount. That's not disbelief in the numbers. It's disbelief in who's on the other side of them.