Nvidia's fiscal second quarter delivered numbers that would have settled most arguments outright. Revenue reached 96.2 billion dollars, more than double what it was a year ago. Data Center revenue, the business that defines the company now, came in at 89 billion dollars, up 117 percent from a year earlier. Adjusted earnings per share hit 2.22 dollars, ahead of the roughly 2.09 dollars analysts expected and up about 111 percent from a year ago.
The stock did not reward that instantly. Shares hovered flat to slightly lower in the first half hour of after hours trading, an odd response to a quarter this size. Over the next two and a half hours, as details from the earnings call spread, the shares climbed steadily to around 219.50 dollars, a gain near 4.7 percent versus the 209.66 dollar regular close. That is not a single noisy print. It built in stages, and it lines up with the timing of the most consequential disclosures on the call: a full year growth forecast well beyond what many investors had modeled, and a major new customer commitment from Amazon Web Services.
The Growth Ceiling Management Insists Isn't There
The most aggressive claim of the night came from Jensen Huang, Nvidia's founder and chief executive, who told analysts on the call that fiscal 2028 revenue growth would run near 70 percent, then added that actual demand is running well above that figure. He framed the gap not as caution about customer appetite but as a hard physical constraint: building out the power, land, and data center shells needed to deploy new compute takes two to three years, regardless of how much a customer wants to buy today.
Pressed by an analyst on how much higher unconstrained demand might be, Huang would only say it would be a lot higher, without giving a number. Another analyst asked him to rank the supply bottlenecks, power availability, memory chips, or foundry wafer capacity, and he declined, saying the entire supply chain is stretched simultaneously and that suppliers across the chain are improving yields daily. That is a confident answer about the direction of demand and a notably vague one about exactly how constrained supply actually is, which leaves the size of the growth ceiling essentially a matter of management's word rather than an independently verifiable number.
Huang also offered a specific economic argument for why the buildout continues: he cited a rise in revenue generated per gigawatt of deployed compute, from roughly 18 billion dollars in the prior Hopper generation to 25 billion dollars for the current Blackwell platform to a projected 40 billion dollars for the upcoming Vera Rubin platform. If that progression holds, each new generation of Nvidia hardware extracts meaningfully more revenue from the same physical infrastructure footprint, which is the core economic case for why customers keep building.
One piece of the growth story remains deliberately walled off. Kress told analysts that China accounted for less than 1 percent of Data Center revenue in the quarter, shipped under export rules governing the Hopper H20 product, and confirmed the third quarter guide assumes zero China data center compute revenue going forward. That means the disclosed base being excluded is small in dollar terms today, but it also means the 108 billion dollar guide is a floor rather than a full accounting of what Nvidia could sell if access to China were restored. Any change in export policy would show up as incremental revenue on top of a guide that currently assumes none, a specific and trackable variable for a market that has otherwise moved on from the China question.
Where the Margin Guide Is Heading
The quarter's least discussed number may be its most important one for the next two quarters. Gross margin held at 75 percent in the just reported quarter, both on a reported and an adjusted basis. Guidance for the current quarter calls for 74 percent, and management pointed to a floor as low as 71 percent by the fourth quarter before an expected recovery toward the low 70s over the following year. Chief Financial Officer Colette Kress attributed part of the pressure to what she called extreme pricing conditions in memory, costs that have run ahead of the company's own internal expectations.
That is a real, quantified headwind that management volunteered rather than one that had to be extracted through questioning, which argues for taking it seriously rather than dismissing it as routine caution. Whether a three to four point margin step down over two quarters is a temporary supply chain wrinkle or the start of a more durable erosion, as component costs rise and competition for the same manufacturing capacity intensifies, is a genuinely open question this quarter does not resolve.
Nvidia Is Now Financing Its Own Demand
Alongside the earnings print, Nvidia disclosed it will commit roughly 18 billion dollars in additional equity investments across the AI ecosystem for the remainder of this fiscal year, part of a broader pattern of capital commitments to the same companies that buy its chips. Management also pointed to more than 500 billion dollars in third party infrastructure financing arranged through partnerships with major asset managers, intended to fund the data centers where Nvidia hardware gets deployed. Separately, Amazon Web Services expanded its relationship with Nvidia to deploy 2 million additional GPUs over the next several years, tripling the scale of a prior commitment.
Huang defended the financing strategy directly on the call, calling the investment opportunity rare and arguing the underlying infrastructure retains its value even if any individual deal does not work out. That is a reasonable argument, and it may well be correct. It is also the kind of statement companies make when they anticipate the strategy will draw scrutiny. The practical question for investors is how much of Nvidia's reported demand is now, directly or indirectly, financed by Nvidia's own balance sheet, and whether that concentration grows from here. Nvidia's own numbers make clear it has the capacity to fund this: the company closed the quarter with roughly 99 billion dollars in cash and short term investments against 39 billion dollars in total debt, generated 24 billion dollars in operating cash flow, and still returned 26 billion dollars to shareholders through buybacks and dividends. Capacity is not the same question as whether the strategy is prudent, and this quarter does not settle that second question either way.
What Still Has to Be Proven
This quarter answered the demand question about as clearly as a single print can. It did not answer the durability question. Nvidia needs to show over the next several quarters that its growing role as a financier of its own customers does not become a source of circularity if AI spending ever slows, and that the current margin step down stays a temporary supply chain issue rather than the start of a structural decline. Investors should watch whether the fourth quarter margin floor holds near the low end management described, whether the pace of the AWS style capacity buildout continues to expand rather than plateau, and whether any of the roughly 500 billion dollars in third party financing commitments show signs of strain. The bull case strengthens if revenue keeps compounding near the rate management described while margins stabilize above the guided floor. The bear case strengthens if margin pressure persists longer than management has signaled, or if the pace of ecosystem financing keeps growing faster than the underlying demand it is meant to support. The burden of proof has moved from whether AI compute demand is real, which this quarter settled, to whether Nvidia can keep supplying and financing that demand without the economics of the arrangement working against it.