The Story
Nvidia has stopped shipping licensed Hopper data centre products to China, interim chief executive Colette Kress told analysts on the company's third-quarter earnings call on 17 November. The company sold $2.4 billion of those products to Chinese customers during the quarter.
Kress said future guidance contains no licensed Hopper data centre products and that the company is no longer shipping them to China. She did not state a reason for the decision, and did not address whether orders had been cancelled or whether inventory charges would follow.
The United States government had approved these sales. The Hopper generation includes the H200, which the administration permitted Nvidia to sell to China. The company's first-quarter filing noted that licences allowing shipment of small quantities of H200 products to specific China-based customers began in February 2026, and that no revenue had been generated under the programme at that point. The restriction described on the call applies to licensed Hopper data centre products; gaming products were not mentioned.
The $2.4 billion represents roughly 2.2 per cent of the quarter's data centre revenue of $107.6 billion. Nvidia guided fourth-quarter revenue to about $130 billion without it.
On 16 November, the day before the results, the US Commerce Department issued rules requiring export licences for AI processor shipments to Indonesia, Malaysia, Singapore, Thailand and Brazil, and restricting the ability of Chinese companies to access advanced AI chips remotely through overseas data centres.
The decision follows the resignation of Jensen Huang on 12 November. Huang had argued publicly and repeatedly against US export controls on advanced chips. The board has ruled out his return and declined to discuss his departure on the call. Nvidia had separately sought approval to sell Blackwell-generation products in China and was not permitted to do so.
Why It Matters
The decision is notable for being voluntary, and cheap, and unexplained, in that order.
Washington had licensed these shipments. The administration approved H200 sales to Chinese customers and Nvidia had begun making them, booking $2.4 billion in a single quarter. Stopping was not compliance with a new prohibition. It was the company declining to use permission it already held, and the interim chief executive offered the call no account of why.
The economics made that easy. Against revenue of $115.8 billion and fourth-quarter guidance of about $130 billion, the abandoned business is roughly 2.2 per cent of data centre sales and about a sixth of the sequential growth the company expects in the next three months alone. A decision that looks like principle is also a decision that costs less than one quarter's increment. That does not make it insincere. It does mean the market will not test it.
The reversal of posture is the larger point. Huang spent years arguing that export controls had failed, that they accelerated Chinese domestic capability, and that the United States should be careful about surrendering the market. He made that case to two administrations. Five days after he left, the company he built voluntarily went further than the controls required, withdrawing from sales the government was willing to license.
Companies do not usually abandon approved revenue without explanation unless the risk of continuing is of a kind they would rather not describe.
The Strategic Read
The timing points at the transshipment problem rather than at China directly.
On 16 November the Commerce Department required export licences for AI processor shipments to Indonesia, Malaysia, Singapore, Thailand and Brazil, and moved against Chinese firms accessing advanced chips remotely through overseas data centres. Those are not China. They are the places through which chips have been alleged to reach China, and remote access through a foreign data centre is the method that makes a physical export control porous.
Megaspeed, the company at the centre of the allegations around Huang's departure, is based in Singapore and has been under investigation over whether it diverted controlled chips to China. Singapore's police opened their inquiry in early 2025. Whatever the outcome, the episode demonstrated that a licensed sale to an approved customer in an approved jurisdiction is only as reliable as the customer, and that the reputational cost of being wrong about one is now borne by the seller.
Read that way, the halt is less about forgoing $2.4 billion of Chinese demand than about declining to administer a licensing regime whose failure modes have just been demonstrated at the cost of a chief executive. A company that cannot be certain where a licensed chip ends up, in a week when Congress is drafting export-control provisions and the Commerce Department is tightening the routes, has an obvious reason to stop and no obligation to explain it.
What the company has not said is whether this is permanent. If it is a pause pending clearer rules, the business returns when the rules arrive. If it is a strategic exit, Nvidia has conceded the Chinese data centre market to domestic suppliers who have had three years of protected demand to improve.
For daily, sharp analysis of the biggest moves in the Indian business and startup ecosystem, follow StartupFox.