The Global Fever

Policy

US Chip Export Restrictions Tighten as Nvidia Halts China H200 Shipments

A new Commerce rule closes the remote-access loophole for advanced AI chips, and Nvidia says it will stop shipping H200s to China. Here is what changes for chipmakers and cloud operators.

Wasif Mumtaz

Founder and Editor · November 19, 2026 · 6 min read

Close-up of a central processor chip mounted on a circuit board
Photo: Magnific

The United States tightened its chip export restrictions on November 16, closing a gap that let Chinese companies reach advanced AI processors through data centers in other countries. Two days later, Nvidia said it will stop shipping its H200 chips to China and guided fourth-quarter revenue from those products to zero. Together, the two moves change how Chinese technology companies can get AI computing power, and they land differently on Nvidia, AMD and cloud operators.

The rule and the shipment halt are separate decisions, but they point the same way. Washington is narrowing the routes by which Chinese firms can buy or rent top-end AI hardware, and Beijing's reluctance to approve imports has left Nvidia with little China revenue to defend.

What the new chip export restrictions require

The Commerce Department's Bureau of Industry and Security issued an interim final rule on Monday, November 16, according to Fortune. It is the first export-control update since a scandal over alleged chip diversion. It targets a long-running gap: a Chinese company could use advanced AI chips without owning them by renting access to servers in overseas data centers.

The rule has two main parts. Exporters now need a license to ship advanced AI chips to data centers in Indonesia, Malaysia, Singapore, Thailand and Brazil. Separately, large data center operators anywhere in the world need a license before they give China-headquartered companies access to computing power.

The rule takes effect once it is published in the Federal Register. Data centers that already serve China-headquartered customers have 90 days from that date to obtain a license. End users that are, or are owned by, U.S.-headquartered companies are exempt. Fortune says Microsoft, Oracle and Amazon likely qualify, although remote-access limits still apply to them.

Fortune also reports that Commerce said the rules were written at the direction of the Treasury Department and the National Security Council. Singapore and Malaysia draw particular attention because Megaspeed, the company accused of diverting chips, is based or has operated data centers there.

How earlier export controls worked, and where they fell short

Until this week, the main barrier was on direct sales. Fortune reports that advanced processors from Nvidia, AMD and Intel are already blocked from sale to Chinese companies. The one exception is the Nvidia H200, which the administration has allowed to be sold to approved Chinese customers.

That exception came with a price. China Daily reports that under an arrangement negotiated by Nvidia co-founder Jensen Huang, H200 shipments passing through the United States face a 25% tariff paid to the U.S. Treasury. The arrangement depended on two governments both saying yes: Washington to license the sale, and Beijing to allow the import.

Direct-sale limits did nothing about rented capacity abroad. A Chinese firm that could not buy a chip could still train a model on one sitting in a data center in another country. The new rule puts that channel under the same kind of license scrutiny as the hardware itself.

Nvidia stops H200 shipments to China

On Wednesday, November 18, Nvidia interim CEO Colette Kress said fourth-quarter guidance excludes all Hopper data-center revenue previously authorized under U.S. approval, according to China Daily. Nvidia sold $2.4 billion of H200 products to China in the third quarter. Revenue from those products is now guided to zero.

The leadership backdrop is unusual. Nvidia announced on November 12 that Huang had stepped down as president, CEO and director with immediate effect, and Kress, the chief financial officer, became interim CEO. China Daily says the company has created a board-level national security committee to oversee export-control compliance, and analysts describe a shift toward a more cautious, compliance-first posture.

Beijing is part of the story. Fortune, citing Reuters, says about 600,000 H200 processors are sitting in Hong Kong while Beijing has largely withheld permission to bring them in. Georgetown researcher Hanna Dohmen suggested that if remote access is cut off, those stranded chips could become more attractive to China.

Jefferies estimates the move could cut Nvidia's data-center revenue by $44 billion to $62 billion next year, China Daily reports. Kress said a large backlog for next-generation products, including the Vera Rubin NVL72 rack-scale system, will cushion the near-term hit.

What the restrictions mean for Nvidia, AMD and cloud suppliers

For Nvidia, the Jefferies range is an analyst estimate, not company guidance. Whether it proves right depends on whether any Chinese demand can be served at all, which in turn depends on Beijing as much as Washington. The company's own message is that its order backlog for newer products limits the short-term damage.

AMD is named alongside Nvidia and Intel among the vendors whose advanced processors are already barred from sale to Chinese companies. The sources reviewed for this article do not give AMD-specific figures for the new rule, so its exposure is not yet clear.

Cloud and data center operators face the most direct compliance work. Operators in the five listed countries need licenses for advanced chip imports, and any large operator serving China-headquartered customers has 90 days to obtain one. The exemption for U.S.-owned end users gives large American providers an easier path than regional operators, though remote-access limits still apply to them.

For Chinese technology companies, the options narrow to licensed access, waiting for Beijing to allow imports, or relying on other hardware. Gartner vice president Roger Sheng said the leadership vacuum at Nvidia has emboldened hardline voices in Washington, and that Asian suppliers and buyers, including Chinese companies, will likely be affected.

What to watch next

Federal Register publication starts the 90-day clock for data centers already serving Chinese customers, so the first compliance deadline falls about three months after the rule takes effect. Watch how many operators apply, and how many instead stop serving China-headquartered clients.

Beijing's decision on the stranded H200s will determine whether Nvidia's China guidance stays at zero. Nvidia's search for a permanent CEO will also show how the company intends to deal with Washington after a change in its top job. Any further Commerce guidance on verifying customer ownership will set how costly compliance becomes for operators.

Frequently asked questions

What do the new US chip export restrictions do?

A November 16 Commerce rule requires licenses to export advanced AI chips to data centers in five countries. It also requires large data center operators anywhere to get a license before giving China-headquartered companies access to computing power. The aim is to stop Chinese firms from reaching restricted chips remotely.

Which countries are covered by the rule?

The rule names Indonesia, Malaysia, Singapore, Thailand and Brazil for the chip-export license requirement. The remote-access requirement for China-headquartered customers applies to large data center operators anywhere.

Has Nvidia stopped selling to China?

According to China Daily, Nvidia will stop shipping H200 chips to China. Interim CEO Colette Kress said fourth-quarter guidance excludes previously authorized Hopper data-center revenue. Nvidia sold $2.4 billion of H200 products to China in the third quarter.

How does this affect AMD?

AMD's advanced processors are already barred from sale to Chinese companies, according to Fortune. The sources reviewed do not quantify the new rule's effect on AMD, so any estimate would be a guess.

Are U.S. cloud companies affected?

End users that are, or are owned by, U.S.-headquartered companies are exempt from the license requirement. Fortune says Microsoft, Oracle and Amazon likely qualify, but remote-access limits still apply to them.

ShareXLinkedInWhatsApp

Sources

Related stories

Policy

APEC Trade Openness in Focus as Leaders Meet in Shenzhen Amid Inflation Risks

Ahead of the Shenzhen leaders' meeting, APEC ministers urged lower barriers and closer policy coordination while the bloc's analysts raised their 2026 inflation forecast to 3.2%.

Policy

ECB December Rate Hike Looms as Bank of England Warns More May Follow

The Bank of England lifted rates to 4.0% and Governor Bailey said more may be needed. With euro-area inflation at 4.0%, the ECB is widely expected to hike in December.

Brands

The Little Shop by Raffles shows luxury hotels moving into retail

Raffles Hotels & Resorts has launched its first global online store, The Little Shop by Raffles, selling fashion, homewares and travel goods beyond the hotel stay.