China moves to curb access to Nvidia H200 chips despite U.S. approval

Beijing authorities are discussing rules that would allow only limited access to the H200 – Nvidia’s second-most advanced AI accelerator – by subjecting purchases to an approval process and requiring buyers to justify why they are not using domestic alternatives, according to people familiar with the talks. A ban on some state-linked entities using the H200 is also under consideration, and no final decision has yet been made.

The Chinese deliberations follow a decision announced on 8 December by the US administration to allow Nvidia to sell H200 chips to approved customers in China. Under the framework outlined by officials, sales would be screened for national security risks and a quarter of the revenue from those exports would be paid to the US government. The authorization does not extend to Nvidia’s newest Blackwell and forthcoming Rubin product lines, which remain restricted.

The H200 had previously fallen under tighter export controls that blocked shipments of many high-end AI processors to China. Those restrictions, introduced over the past three years, hit Nvidia’s ability to grow in one of the world’s largest AI hardware markets and pushed the company to design weaker, China-specific chips such as the H20.

Regulators in Beijing are now discussing how to handle the H200 within China’s broader industrial policy. Options on the table include requiring companies to obtain explicit authorization before importing the chip and to explain why local processors are not sufficient for their AI workloads. Officials are also examining whether to restrict listed or state-controlled companies from buying the H200 at all, in order to steer public-sector and strategic projects towards Chinese suppliers.

The debate reflects two parallel goals: maintaining access to leading-edge hardware for domestic technology groups that still rely on foreign accelerators, and reducing long-term dependence on US products after several rounds of export controls. Over the past year, Chinese authorities have already pressured firms to scale back purchases of Nvidia’s H20 and accelerated support for homegrown AI chips, prompting Nvidia to halt H20 production earlier in 2025.

The H200 is an upgraded successor to Nvidia’s widely used H100, designed for training and running large language models and other compute-intensive AI systems. It is not the company’s most powerful chip – that status now belongs to the Blackwell family – but it remains a high-end part that many Chinese cloud and internet companies have sought to secure for data centres and AI research.

Because US rules still block exports of Nvidia’s top-tier products, the H200 represents a critical middle ground: significantly more capable than older hardware, yet permitted for export under specified conditions. A Chinese regime that limits or channels access to the H200 could reduce the commercial upside of the US decision while giving domestic chipmakers more room to compete.

Nvidia’s shares initially rose after the US announced it would allow H200 exports to China, reflecting expectations that the company could recover part of the sales lost to earlier restrictions. Gains narrowed once reports emerged that Beijing was considering its own curbs, and the stock ended the day only modestly higher. Other large US chipmakers, including AMD and Intel, also recorded small advances amid speculation about future export permissions.

In Washington, resistance to broader sales continues. Lawmakers from both parties have backed proposals that would delay or block exports of advanced AI chips to China for up to 30 months, citing concerns that such hardware could strengthen Chinese military and cyber capabilities. Those efforts sit alongside export-control measures first rolled out in October 2022 that targeted advanced computing and semiconductor manufacturing items bound for China.

China’s largest internet and cloud companies remain caught between policy pressure and technical requirements. Many have sought to keep using Nvidia hardware by renting capacity or training models in overseas data centres where export rules permit deployment, while also testing local alternatives at home. Access to the H200 inside China would ease performance bottlenecks for some of these firms, but stricter domestic controls could push more of their cutting-edge work offshore or onto domestic chips.

At the same time, Chinese industrial policy has increasingly prioritised semiconductor self-sufficiency. Authorities have directed subsidies, financing and procurement preferences toward domestic AI chip producers, and have already warned some companies against buying certain foreign processors. The emerging approach to the H200 fits into that pattern of encouraging local alternatives while keeping a narrow channel open to foreign technology under tight oversight.

The latest developments are part of a broader stand-off over advanced chips and AI infrastructure between the world’s two largest economies. Since 2022, the US has progressively tightened rules on the export of high-end GPUs and chip-making tools to China, arguing that cutting-edge AI hardware can be used for military applications and surveillance. China has responded by increasing investment in its own semiconductor sector, imposing restrictions on certain foreign products and looking for ways to reduce supply-chain vulnerabilities.

Against that backdrop, the H200 export decision and China’s planned internal limits illustrate how both sides are now calibrating access at a more granular level – not only deciding whether chips can cross borders, but also which institutions and projects can use them once they arrive. How those rules are finalized will help determine the shape of the AI hardware landscape in China and the size of the addressable market for Nvidia and its US peers.