Inside Nvidia's Record $108 Billion Forecast: A Fourth Quarter of Deceleration

Nvidia answered the market's biggest question on Wednesday with a number that should have settled it. Revenue for the fiscal second quarter, ended July 26, came in at $96.2 billion, up 106% from a year earlier and roughly $4 billion above the consensus analysts carried into the report. GAAP earnings were $2.46 a share. Data-center sales — nine of every ten dollars of the business — reached $89.0 billion, up 117%. And the company guided to $108.0 billion for the current quarter, plus or minus 2%, roughly $4 billion above the average analyst forecast.

The shares slipped in extended trading anyway. For anyone trying to judge the AI trade, this was the quarterly referendum, and the market's reaction said the report was not as reassuring as it looks. Nvidia's sales forecast was never the problem. Its shape was.

For a growth stock, momentum means the rate at which sales grow — whether the next quarter speeds up or slows down — and that rate is now visibly bending. Total revenue rose 20% in the quarter that ended in January and 20% again in April; the July quarter grew 18%, and the new forecast implies about 12% for October. That is four straight quarters of sequential deceleration. The dollars being added each quarter have stopped climbing too: $14.6 billion between April and July, and roughly $12 billion built into the October forecast. Record level, cooling rate — the two can coexist, and right now both are true.

The year-over-year view looks less steep than the sequential one, and that gap is itself part of the story. The reported 106% growth ran against a soft comparison: a year earlier, sales of $46.7 billion were depressed by the export ban that cut off sales of its H20 chip to China entirely. The $108 billion forecast still implies nearly 90% growth over the same quarter a year ago — deceleration, but less dramatic than it looks. The sequential slowdown is the cleaner number, because it does not depend on what an old quarter looked like.

Which is why the two halves of Wednesday's release sit in tension. Jensen Huang's prepared remarks say demand is "accelerating" and that "the AI infrastructure buildout is at full steam," with Vera Rubin, the next platform, "now in full production." A forecast that slows while its CEO describes accelerating demand is exactly the contradiction the market is trying to price.

Part of the slowdown is Nvidia-made and would slow the guide regardless of demand. The company built its $108 billion outlook with zero data-center chip sales to China, a market shut off by export policy for most of the past year; even a partial policy reversal could add revenue above the guided number. The quarter also straddles the handoff from Blackwell to Vera Rubin, the kind of product transition that absorbs a few points of sequential growth even when orders are strong. And Nvidia has now beaten its own guidance in thirteen consecutive quarters, so the published number functions as a floor — something the market already knows and bakes in before the stock can react.

A deceleration that is partly by construction is not the same as a collapse in demand. But it does not resolve the question investors keep failing to get comfortable with: whether the spending at the top of the chain still has enough momentum to feed Nvidia's ramp. Microsoft and Meta, among others, are projected to spend more than $730 billion on AI infrastructure this year, up from about $400 billion last year. Nvidia's guide is the cleanest existing translation of that upstream budget into chips leaving the factory, which is why a cooling sequential curve gets read as the first sign that the capex program is maturing. A rising share of that budget is also going to the cloud giants' own in-house silicon, designed to reduce how much lands on Nvidia.

The most consequential part of this forecast is not in the forecast at all. Two weeks before earnings, Nvidia signed memorandums of understanding with six financial institutions — Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR — to build platforms that would mobilize over $500 billion of third-party capital for AI infrastructure, converting Nvidia's compute into an asset class that pension and insurance money can own, with long-duration, usage-linked revenue. Huang's summary was five words: "In AI, compute is revenue."

That is the structural answer to the momentum question. If AI data centers become a financed, yield-bearing asset class, demand stops depending on the four hyperscaler balance sheets; capital from outside the industry keeps the order book filled. It is also the risk, concentrated in one line. Reuters reports Nvidia has agreed to guarantee deals under the program, and analysts have already counted on the order of $300 billion in equity stakes, guarantees and backstops tied to its customer financing. Some of the demand behind the record numbers is being underwritten by Nvidia itself. If the asset class works, this is how a $108 billion quarter becomes a signpost rather than a peak. If the financing sours, revenue growth and credit exposure will rise together.

A good guide and a cooling curve are not contradictory; they have coexisted since the January quarter. What would separate a pause from a peak is concrete and observable: whether the $500 billion of memorandums convert into funded, committed deals over the next two quarters; whether hyperscalers raise — not merely repeat — their capital budgets when they set 2027 plans; whether gross margin holds near the 75% the company has guided to while Vera Rubin ramps; and whether policy reopens the Chinese market the current guide assumes away. Nvidia's forecast is a floor, and a strong one. The rate inside it is the part that still has something to prove.