The last full quarter of H200 sales to China was worth $2.4 billion. Nvidia closed the channel one day before it reported the number.
That order of events matters. The company's interim CEO told shareholders on November 17 that it would no longer supply licensed Hopper data center products to China, which means the H200s cleared to ship under the US arrangement stop going out. The next afternoon, the quarterly report landed with the Chinese revenue still inside it.
A small line in a very large quarter
The quarter ended October 25. Revenue was $115.8 billion, up 103% from a year earlier and 20% from the prior quarter. Consensus was $109.33 billion, and the company's own guidance midpoint was $108 billion. It was the 17th consecutive quarter Nvidia beat the revenue estimate.
GAAP net income was $65.287 billion, up 105%. Diluted earnings per share came to $2.70. Non-GAAP gross margin was 74.9%, against expectations of about 74%.
Data center revenue was $107.6 billion, up 110% year over year and 21% sequentially, an increase of $18.6 billion. That one segment was about 93% of the whole company.
Two percent of everything
Set the $2.4 billion against those numbers and the China exit looks almost free. It is roughly 2% of quarterly revenue and a little over 2% of data center sales.
Almost free is not the same as free.
What Nvidia gave up was not volume so much as a legal channel, the only route by which a permitted part could reach Chinese buyers. Revenue can be replaced. A channel, once closed, has to be renegotiated.
The growth that covered the gap came from somewhere else. Hyperscale customers bought $60.1 billion, up $11.4 billion sequentially, or 23%, against expectations of $53.3 billion. The AI cloud, industrial and enterprise group, known as ACIE and including sovereign customers, bought $47.5 billion, up $7.2 billion, or 18%, against expectations of $48.9 billion. The two add up to the $107.6 billion data center total.
What Beijing said, in full
An Agence France-Presse correspondent raised the halt at the foreign ministry briefing on November 18. Spokesperson Guo Jiakun answered with one sentence: "China has already stated its principled position on the US chip export issue on multiple occasions."
That is the entire official response as recorded. It contains no new measure and no new commitment. Read against the A-share tape the same day, it did not need to.
What Shanghai bought instead
The GPU index rose more than 2% that session, with the domestic accelerator names leading. Which ones, and by how much, is a story of its own. What matters here is the direction of the trade: money moved toward the chips that would replace the one that stopped shipping.
That is a substitution trade, and it is pricing a product that has not yet shipped at scale. The clearest statement of the near-term cost came from an operator rather than a broker. Tencent executives, discussing the company's third-quarter results, acknowledged that available compute is falling and that short-term model training has been affected.
The board saw it coming
Nvidia's board has created a national security and export controls oversight committee, drawing on independent directors, senior external advisers and legal counsel. In the same quarter, purchase commitments rose to $390 billion, more than three times quarterly revenue. Of that, $32 billion is tied to agreements with emerging cloud providers, and total guarantees stood at $118 billion at period end.
A company that signs supply commitments at that scale while standing up a board committee for export risk is telling you where it thinks the constraint sits. Not in demand. In permission.
The next number
Guidance for the current quarter is about $130 billion, plus or minus 3%, with gross margin of 71% to 72% on both GAAP and non-GAAP bases. That is roughly three points below what Nvidia just delivered, and still a margin most hardware companies never reach. Management also raised its fiscal 2028 revenue growth expectation to about 75%.
The cloud industry backlog it tracks went from $2 trillion to $2.5 trillion in three months. Vera Rubin, in its first quarter of volume production, contributed close to $20 billion, about 18% of data center revenue. Networking passed $21 billion, about a fifth of the segment, with co-packaged optics scale-out products in volume production.
None of that depends on China.
All of it depends on the same handful of buyers continuing to spend, and on export rules that now change faster than a product cycle.
The H200 decision removed about 2% of revenue and the last direct argument Nvidia had with Washington. The bill for that is small. The precedent is not.
*Disclaimer: This article is based on publicly available information and is for reference only. It does not constitute investment advice.*
