Nvidia Stops Selling H200 Chips in China as Export Rules Tighten

Nvidia will no longer sell its H200 chips in China, acting CEO Colette Kress said on the company's first earnings call since co-founder Jensen Huang stepped down the week before. The statement, reported by TechCrunch, reverses an arrangement Huang had helped put in place with President Donald Trump.
The same day, Nvidia said it is creating a board-level committee to oversee national security policy, export control compliance and global trade risk. The committee is at the planning stage, and its membership and mandate beyond those three areas were not part of the company's statement.
Under the H200 deal, chips destined for China had to be shipped through the United States, which allowed the government to collect a 25% tariff on them. China had largely refused to accept H200 deliveries even before Nvidia decided to stop shipping them, so the commercial effect of the reversal is smaller than the headline figures alone would suggest.
The regulatory change behind the decision arrived a day earlier. New U.S. rules issued on November 16, 2026 bar advanced U.S. chips from being used by foreign data centers to serve Chinese customers remotely. Large data centers that want to buy Nvidia chips will have to certify that they are not serving any Chinese customers. The rules reach beyond the physical location of a chip. They govern who is on the other end of a remote session, so the rule follows the renter rather than the parking spot, and a cloud provider's compliance position now depends on its tenant base as well as its hardware footprint.
The financial exposure is the number most analysts will watch. Investment bank Jefferies forecast that the new export rules might trim between $44 billion and $62 billion from Nvidia's data center revenue next year. Kress offered a different read. She said backlog demand for advanced chips such as Vera Rubin, Nvidia's next-generation data center platform, means the new rules would have no impact on the company's top-line revenue.
Nvidia's own disclosures add a complication. Its Singapore subsidiary reached annual revenue of $104 billion in 2025, and Nvidia stopped disclosing figures for that subsidiary after that year. A large share of the company's sales therefore flows through a jurisdiction whose reported revenue is no longer visible to outside observers at the same granularity.
The gap between the Jefferies range and Kress's no-impact statement comes down to one question: whether the order backlog can absorb the volume that will no longer move into China. The answer will show up in quarterly figures over the coming reporting periods, not in the forecasts issued this week. For the rest of the industry, the more immediate work is in the compliance details. Data center operators, colocation providers and cloud platforms will need to document their customer base to satisfy the certification requirement, a process with its own engineering and legal cost.
The broader context here is that the chip supply chain is now tightly bound to policy. A deal brokered at the head of government, a tariff collected at the border, a rule reaching into remote compute access, and a board committee created to manage all of it are separate instruments. Together they make export control a standing feature of a vendor's operating model rather than an occasional licensing question.
In my view, the more useful question for engineers and buyers is not whether Nvidia's top line holds, but how quickly the supply headed for China gets redirected and where that capacity lands. Accelerator demand that cannot be served in one market tends to find another buyer, and the speed of that redirection will shape pricing and allocation for everyone else competing for the same silicon. Clearer rules, if applied consistently, also give infrastructure teams a defined boundary to build against, which is a more workable position than uncertainty over what counts as a permitted customer.


