Nvidia is expected to report quarterly revenue of approximately $92 billion when it announces results Wednesday, roughly 100 percent above the same period a year ago, and above management's own guidance of $91 billion. The tension heading into the print is that investors already price in a beat. What they do not yet know is whether the gross margin holds.
The margin as the signal
Motley Fool contributors Matt Frankel and Rachel Warren, speaking on August 24, walked through why the gross margin is the number that does the work. Nvidia's margin has climbed from roughly 58 percent a decade ago to approximately 74 percent today. Frankel attributed that to Nvidia's dominant position: the company holds roughly 95 percent of the data center GPU market and has been essentially sold out of chips for data centers for the coming years, giving it what he described as total pricing power.
Warren added that the hyperscalers, Microsoft, Amazon, and Alphabet, are ordering next-generation chips faster than Nvidia's manufacturing partner TSMC can produce them. Nvidia passes rising input costs, including surging prices of high-bandwidth memory from SK Hynix, directly to customers without losing order volumes. The lock-in runs deeper than supply constraints. Nvidia's proprietary CUDA software ecosystem means optimizing an AI model for a competing architecture takes months of engineering work, and the tech giants choose to pay Nvidia's premium rather than risk falling behind.
Warren identified 70 percent as the gross margin floor she would watch closely. A slide toward that level could signal that supply has caught up with demand, or that AMD's MI 300 series or the hyperscalers' own custom chips have achieved enough software compatibility to erode Nvidia's pricing position. AMD recently launched its first full-scale rack system for data centers, which is what makes the margin print consequential beyond one quarter.
The counterargument
The efficiency-shrinks-the-market thesis is the strongest pushback on the bull case. Warren framed it through the Jevons Paradox, the observation by 19th-century economist William Stanley Jevons that when steam engines became more fuel-efficient, coal consumption skyrocketed because cheaper power opened entirely new industries. Applied to AI data centers, cheaper inference could create waves of demand that absorb efficiency gains rather than reduce the physical footprint.
Jon Quast raised the harder version of the bear case: a significant portion of today's AI demand may be subsidized, and subsidized demand has a finite lifespan. Data center vacancy is running at roughly 1 percent against a historical norm closer to 5 percent, which argues the present supply-demand balance is tight. Whether that holds as competition scales is precisely what the margin line will begin to answer.
On balance, the earnings move far more than Nvidia's own stock. Frankel listed Cisco, Arista Networks, and Applied Materials as companies whose demand outlook shifts with whatever Jensen Huang says about the pace of the AI infrastructure build-out. The line to watch is the gross margin, and any commentary from Huang on hyperscaler capital expenditure heading into next year.