Nvidia's Q2 revenue hit a record $96.2 billion, up 106% year over year -- its stock fell on a memory-cost margin warning, then rose 6% on a single guidance number

Nvidia reported $96.2 billion in revenue for the quarter ended July 26 -- up 106% from a year ago and comfortably ahead of the roughly $92.2 billion Wall Street had modeled -- and then spent most of its earnings call talking about the number that wasn't a beat. Gross margin held at 75.0% this quarter but is guided to keep sliding through Q4, and CFO Colette Kress told investors memory costs have "exceeded our prior expectations and are headed even higher into next year." Nvidia shares fell on that warning during the call, then reversed to close up roughly 6% once CEO Jensen Huang reframed the quarter's other headline number -- a 70% FY2028 growth guide -- as a ceiling set by chip supply, not customer demand.

The revenue itself was unambiguous. Data Center revenue hit a record $89.0 billion, up 117% year over year, against analyst estimates of roughly $85.7 billion. Nvidia's newer reporting split breaks that into $48.7 billion of hyperscale revenue, up 13% sequentially, and $40.3 billion from AI clouds, industrial and enterprise (ACIE) customers -- sovereign AI programs, AI-native startups, and on-premises enterprise buyers -- which grew 25% sequentially, nearly twice hyperscale's pace. The remaining $7.2 billion came from Edge Computing, up 13% sequentially. Non-GAAP earnings per share came in at $2.22, ahead of the roughly $2.06-$2.09 analysts had modeled; GAAP EPS was $2.46.

Guidance for the current quarter is $108.0 billion in revenue, plus or minus 2% -- 18% higher than the quarter just reported, if hit. That figure comes with an explicit exclusion: Nvidia states it is not assuming any Data Center compute revenue from China in the outlook at all, citing continued uncertainty over export and import licenses. Gross margin guidance for the new quarter is 74.0%, plus or minus 50 basis points -- the first step down a path the company says continues to roughly 71-72% by the fiscal fourth quarter, entirely on the cost side rather than pricing.

The cause, on Kress's own account, is a memory market Nvidia's own buildout has helped tighten. High-bandwidth memory and broader memory-component prices have risen faster than the company modeled even a quarter ago, and Nvidia's own supply commitments -- tied largely to Vera Rubin production -- have swelled into the hundreds of billions of dollars. Days sales outstanding, a measure of how long Nvidia waits to get paid, was extended to 60 days for its largest customers -- a change that shows up as looser financing terms for the buyers underwriting this build-out, not as a margin problem Nvidia is absorbing alone.

The FY2028 growth figure Huang chose to lead with is deliberately a supply number, not a demand ceiling. "Even though our demand is much greater than 70%, our supply allows us to confidently deliver 70%," he told analysts, declining to quantify the gap beyond calling unconstrained demand "a lot higher." "Compute is revenue," Huang said elsewhere on the call, framing the quarter as evidence that AI "has reached its inflection point" and that its outputs are now "productive and profitable" rather than speculative. It's a framing investors have heard from Nvidia before; what moved the stock this time is that it arrived paired with a specific, dated margin decline instead of an unqualified growth number standing alone.

The reaction wasn't confined to Nvidia's own stock. AMD and Intel shares also ticked up alongside Nvidia's after-hours move, per the same earnings-day coverage -- a sign investors read the 70% growth guide as a signal about AI infrastructure spending broadly, not just Nvidia's own order book, even though the margin story that started the session's selling was specific to Nvidia's own memory-cost exposure.

Nvidia returned $26.0 billion to shareholders this quarter through buybacks and dividends, with roughly $99.0 billion left on its repurchase authorization -- capital return at a scale that itself depends on the margin holding somewhere near its guided floor. The result lands in the same stretch of weeks Nvidia has been signing deals that route its own growth through other companies' balance sheets: a warrant worth up to $12.2 billion tied to Google's custom-chip purchases, reported investment talks with Perplexity at a $30 billion-plus valuation, and a compute license with Poolside. This quarter's headline number says the core GPU business doesn't need those deals to grow; the margin guidance says growing it is getting more expensive to supply either way.