Nvidia reports its second quarter of fiscal 2027 after the US close on Wednesday, August 26. The consensus is $91.85 billion of revenue, against a company guide of $91.0 billion, plus or minus 2%.
Those two numbers are 0.93% apart. The Street is sitting inside Nvidia's own tolerance band, which runs from $89.18B to $92.82B. Whatever Wednesday's revenue line does, it is extremely unlikely to be news: the company told everyone in May, and everyone agreed.
That is worth stating plainly because the alternative framing — a company about to grow revenue 96.5% year on year — sounds like it should be the story and is not. Growth of that size is what the guide already promised. The surprise, if there is one, has to come from somewhere else.
The panels above carry both consensus lines, and the chart under them is the one that matters for grading: NVIDIA has beaten the estimate in each of the last four quarters, by 5.1%, 5.3%, 8.3% and 10.0% — a run of beats that has been getting wider, not narrower.
One caution before any of those numbers is used. The EPS consensus depends on which feed you read: $2.01 across 11 estimates, $2.08 across 40. That 3.4% spread is not a rounding difference — it is a third of the smallest beat in the run above, and it decides whether an identical print is reported as a beat or as in line. The four surprises quoted here and the $2.01 are on one provider's basis, which is what makes them comparable to each other; that same provider puts the July 2025 quarter at $0.99 where NVIDIA's own release says $1.05 non-GAAP.
And NVIDIA's non-GAAP EPS runs below its GAAP number — $1.87 against $2.39 in the April quarter — which is the reverse of most companies. Anyone comparing Wednesday's non-GAAP print to a GAAP prior will manufacture a miss out of an accounting convention.
The chart above the article carries the same comparison quarter by quarter: four graded quarters, average surprise +7%, then the three older prints we hold reported figures for but no matched estimate, and finally Wednesday's $2.01 sitting alone with nothing beside it yet.
The three lines that are actually open
1. The October guide, against $103.96 billion. This is the trade and everyone knows it. Consensus for the guide is $103.96B; at least one analyst models a guide near $108B. Nvidia has cleared its own revenue guide every quarter through this ramp, so the October number is where management's read on Rubin-generation demand becomes visible. A guide below roughly $104B does not require a demand problem to hurt — it only requires the absence of an acceleration the price already contains.
2. Gross margin, where two credible parties disagree by two points. Nvidia guided 75.0% non-GAAP, ±50 basis points. Independent modelling puts it at 73–74% on rising memory cost and early Rubin yields. That is the first genuinely two-sided debate on this line in eight quarters, and it is not abstract to us: we have covered SK Hynix tripling revenue in a year at a 76.3% operating margin, and Micron's gross margin going from 37.7% to 84.6% on high-bandwidth memory. Nvidia's input cost is somebody else's record margin. Both sides of that trade are on this site already, and only one of them can be right about where the value settles.
3. Whether Rubin pricing holds. Our Nvidia model prices the data-centre compute line on NVL72-class racks at $3.0 million a unit, drifting down 1% a quarter. Vera Rubin is the generation that replaces them and is in production now — CoreWeave's own quarter described the "first bring-up and validation of NVIDIA Vera Rubin NVL72". A guide built on Rubin pricing is a direct test of that $3.0 million assumption. It is ours, not a Nvidia disclosure, and Wednesday is the first quarter that can move it.
The claim we made in February, and what four quarters have done to it
Our February preview made a specific argument: "The problem isn't the numbers. The numbers are absurd. The problem is expectations have gotten so high that beating them doesn't move the needle anymore."
We graded that call on August 15 and split it deliberately. The business half settled against us: Q4 revenue came in at $68.127B against the $65.6B consensus we named, EPS at $1.76 against $1.52, and the following quarter at $81.615B. The numbers did not disappoint. The reaction half we said this site could not settle, and left it there.
It is closer to settled now, and it is worth being precise about how much closer. Nvidia has beaten consensus in each of the last four quarters, and the stock has fallen after each one. That direction is corroborated across independent accounts of those days: the February print drew the largest decline, the May print a smaller one, the two before them smaller still.
We are not going to print four decimal-place percentages for those moves. They are quoted inconsistently — published figures for the February reaction alone range from about 4% to 5% depending on whether the source measures the close, the next session or the intraday low, and this site holds no historical daily price series to arbitrate. What survives verification is the sign, four times running, not the magnitude.
Which means the honest version of our February claim is narrower than the way we wrote it: four beats have produced four declines, and four is a small number. Three of the four moves were modest enough that "the market shrugged" describes them better than "the market sold off". A fifth data point arrives on Wednesday. That is enough to make the pattern worth naming and not enough to make it a law, and we would rather say so before the print than after it.
What the price already contains
At the $216.85 close on August 20, Nvidia trades around 24× forward earnings — roughly 20% below its 52-week high. Our own model's base case is $245.07, and our China-returns case is $350.31. The Street's average target across 62 analysts is $304.73, on a range from $180 to $500.
That is the useful frame: the Street's target sits inside our own spread, and almost the entire distance between our base case and our China case is China. Our H200 piece established that the outlook assumes zero China data-centre compute revenue, that the 20,000 H200 units Beijing has acknowledged are worth $0.5–0.8B — inside the ±2% tolerance on the guide, so invisible in it — and that rebuilding the old $4.6B China quarter would take roughly 140,000 units a quarter. Nvidia has since publicly denied a report of a China-specific inference part.
China is therefore unpriced optionality rather than a swing factor for Wednesday. Nothing in this quarter's guide depends on it, and nothing management says about it on the call changes the reported number.
What to watch
- The October guide against $103.96B. Above it, the acceleration the price assumes is intact. Below it, no demand problem is required for the stock to fall.
- Non-GAAP gross margin against the 75.0% guide. The two-point spread between the company's band and independent estimates is the widest disagreement on this line in two years, and memory cost is the mechanism.
- Any Rubin pricing disclosure, against the $3.0 million per NVL72-class rack our model assumes. This is our number, and it is the one this print can most directly falsify.
- The Hyperscale and ACIE split inside data centre. If the growth baton is genuinely passing from the hyperscalers to everyone else, it shows up here first, and it changes who the customer concentration risk belongs to.
- The reaction itself. Fifth beat, fifth decline would make the pattern harder to dismiss. A rally on an in-line quarter would end it.
Sources and provenance. The August 26 report date and the after-close timing are NVIDIA's own. Revenue guidance of $91.0 billion ±2%, GAAP gross margin of 74.9% and non-GAAP of 75.0% each ±50 basis points, and the first-quarter actuals of $81.615 billion of revenue, $75.2 billion of data centre revenue, $2.39 GAAP and $1.87 non-GAAP diluted EPS are all from NVIDIA's first-quarter fiscal 2027 release of May 20, 2026. The year-ago July quarter figures of $46.74 billion and $1.05 non-GAAP are from the corresponding release. Consensus of $91.85 billion of revenue, the $103.96 billion October guide consensus and the $304.73 average target across 62 analysts are press-reported from third-party estimate feeds; this site stores no consensus series for NVIDIA and does not verify them. The EPS consensus of $2.01 and the four prior-quarter estimate/actual pairs — $0.94 against $0.99, $1.18 against $1.24, $1.45 against $1.57 and $1.70 against $1.87 — are one estimate provider's figures, taken as a matched set so the surprises are internally comparable; a broader 40-estimate feed carries $2.08 for this quarter, and the same provider puts the July 2025 quarter at $0.99 where NVIDIA reported $1.05 non-GAAP. The 73–74% gross-margin estimate and the modelling of a guide near $108 billion are individual analysts' published work, relayed through third-party summaries, and are attributed rather than endorsed. The direction of the last four post-earnings share-price moves is corroborated across independent press accounts; the magnitudes are quoted inconsistently between sources and this site holds no historical daily price series, which is why no percentage is printed here. The $216.85 price is the August 20 close and will not be the number you see today. The $245.07 base case, the $350.31 China-returns case and the $3.0 million NVL72-class rack price with 1% quarterly drift are assumptions in our own NVIDIA model, not company disclosures. The Vera Rubin NVL72 bring-up quote is from CoreWeave's own quarterly release. SK Hynix and Micron margin figures are from those companies' own reported results.
