NVIDIA (NASDAQ: NVDA) delivered a third quarter that was less a beat and more a confirmation of trajectory. Revenue of $115.8 billion surged 103% from a year ago and 20% sequentially, topping the $108.8 billion consensus by $7 billion. Non-GAAP diluted EPS of $2.68 came in $0.21 above estimates.
The headline numbers are striking. The real signal lies in what they reveal about the Vera Rubin transition and the cost of scaling at this velocity.

Data Center revenue hit $107.6 billion, up 110% year-over-year and 21% sequentially. The AI Clouds, Industrial, & Enterprise segment nearly tripled from a year ago to $47.5 billion. Hyperscale grew a more measured 71% to $60.1 billion. The Vera Rubin NVL72, which NVIDIA calls the fastest product ramp in company history, is now live at CoreWeave, Google Cloud, Microsoft Azure, Oracle, and Nebius. The company also announced an agreement to acquire Hugging Face, a move into open AI platform capabilities.
Gross margin tells a more nuanced story. GAAP gross margin of 74.8% was up 140 basis points year-over-year but down 20 basis points sequentially. Non-GAAP gross margin of 74.9% followed the same pattern. The sequential dip reflects the initial Rubin ramp, where new architectures carry higher costs before scale benefits materialize.
That dynamic intensifies in Q4. Guidance calls for GAAP and non-GAAP gross margins of 71.0% to 72.0%. A 300-basis-point sequential compression at the midpoint is not a margin problem. It is a mix story. As Rubin becomes a larger share of shipments, gross margin will compress before it expands. The question is how long that compression lasts.
Q4 revenue guidance of $130 billion plus or minus 3% implies roughly 12% sequential growth at the midpoint, even with an extra week in the quarter. The company explicitly assumes no Data Center compute revenue from China. The guide is built entirely on demand from the rest of the world. That is a statement of confidence in non-China demand. It also means any change in the China export environment would be a one-way risk to the upside, not the downside.
Capital allocation was aggressive. NVIDIA returned a record $28.1 billion to shareholders in Q3, including $22.1 billion in buybacks and $6.0 billion in dividends. The board added $150 billion to the repurchase authorization, bringing the total remaining to $227.3 billion, with plans to execute through fiscal 2028. The buyback pace looks aggressive given the scale of the infrastructure commitments the company is making. Supply commitments jumped to $390 billion from $279 billion last quarter, driven by multi-year memory procurement. Additional guarantees for land, power, and shell reached $118 billion, including a $105 billion guarantee tied to SB Energy's Ohio campus that will exclusively host NVIDIA infrastructure for OpenAI under 20-year leases. The company is placing a series of large, long-duration bets on the durability of AI demand.
Edge Computing revenue of $8.2 billion grew 42% year-over-year and 14% sequentially, driven by Blackwell workstation sales and Physical AI applications. Consumer PC demand remained constrained by higher memory and system prices. The enterprise and robotics side of the edge business is building momentum.
The forward read is straightforward. NVIDIA is executing on a product transition that is both faster and larger than any in its history. The gross margin compression in Q4 is a function of mix, not competitive pressure or cost overruns. The real risk is not demand. It is execution at scale. Supply commitments of $390 billion, cloud service agreements and data center leases of $102 billion, and land and power guarantees of $118 billion represent a balance sheet increasingly leveraged to the assumption that AI infrastructure demand compounds for years. If that assumption holds, the current margin compression is a temporary feature of a durable growth story. If it falters, the commitments become a weight. For now, the data supports the first scenario.
