US relaxes export rules for Nvidia’s H200 AI chip amid Chinese import restrictions and domestic push for semiconductor independence

The US has revised its export policies to permit conditional sales of Nvidia’s H200 AI accelerator to China, while Beijing tightens restrictions on imports, accelerating China’s domestic chip development and reshaping the global AI hardware market.

The United States has formally shifted its export policy to allow conditional sales of Nvidia’s H200 artificial-intelligence accelerator to China, while Beijing has moved to tightly restrict imports of the same hardware, effectively keeping the chip out of most commercial Chinese use. According to the U.S. Federal Register notice published by the Commerce Department’s Bureau of Industry and Security, H200 and equivalent products will be subject to individual licence reviews rather than the previous presumption-of-denial approach. The Chosun Ilbo reported the change and described Beijing’s immediate response of allowing H200 imports only in “special cases” as part of a drive for technological self-reliance. ^[1][2]

The Commerce Department’s new licensing framework imposes detailed conditions intended to limit risk and preserve U.S. supply. Industry reporting outlines requirements including prioritisation of domestic U.S. demand, independent performance testing by U.S.-based labs, strict know-your-customer checks and limits on the volume of chips that can be sent to China relative to U.S. deliveries. Tom’s Hardware summarised the rules as capping Chinese-bound shipments to no more than 50% of what is supplied to the U.S. market and restricting exports to accelerators beneath specified performance thresholds, while AP News noted additional constraints such as prohibitions on military use. These measures effectively treat China as a secondary market and erect compliance burdens likely to limit large-scale exports. ^[4][2]

Beijing’s countermeasures are immediate and precautionary. Reporting from Tom’s Hardware and The Information indicates Chinese authorities have instructed major firms to pause or restrict H200 orders and are restricting approvals to “special circumstances,” reportedly focused on university research and development labs. The Chosun coverage quoted The Information’s analysis that ‘Chinese authorities have not clearly explained specific guidelines or permissible scopes for the “necessary cases” mentioned in the guidelines,’ suggesting the rule’s vagueness preserves government discretion and leaves room for future easing if bilateral relations improve. Some accounts indicate Beijing considered requiring purchasers of H200s to buy a proportionate share of domestically produced chips, but has signalled even stricter import controls in practice. ^[3][1]

The move carries mixed commercial consequences for Nvidia and the broader semiconductor trade. Nvidia welcomed the U.S. decision in market commentary, while industry analysts and legislators have flagged national-security concerns about even limited sales. Axios reported the Trump administration has also imposed a 25% tariff on selected semiconductors sold to approved Chinese commercial customers, a step intended to both raise revenue and constrain transfers of advanced compute capability. Forbes analysis argues that formal, regulated export channels and associated fees will improve revenue visibility for Nvidia by shifting demand into licenced, transparent sales, even if volumes remain constrained by compliance rules and Chinese limits. ^[2][6][7]

Longer term, China’s policy and domestic progress signal a rapid market realignment. Analysts cited by Tom’s Hardware and market research from Bernstein project that Nvidia’s share of China’s AI accelerator market could collapse from the high double digits in 2024 to a single-digit share by 2026, as domestic vendors such as Huawei, Moore Threads, MetaX and others expand capacity and product breadth. Those firms are closing the gap on older Nvidia architectures, though they still trail the very top-end Blackwell chips in performance; Chinese strategy links market protection with a five-year push for semiconductor independence and scaling of more advanced foundry processes. Industry observers note that even as imports of H200 are permitted in narrowly defined cases, China’s broader industrial policy, combined with U.S. export constraints, will accelerate the transition to a largely domestic AI hardware stack. ^[5]

For now, the net effect is a tightly managed pathway rather than a reopening of unfettered trade. U.S. rules create a licensed corridor for lower-tier high-performance accelerators under strict conditions, while China’s “special cases” approach and broader incentives for local suppliers limit adoption by major cloud providers and hyperscalers. The outcome leaves Nvidia able to sell into China only in constrained, compliance-heavy circumstances and accelerates Beijing’s push to replace imported compute with homegrown alternatives, keeping geopolitical and commercial uncertainty at the centre of global AI infrastructure planning. ^[4][1][5]

📌 Reference Map:

##Reference Map:

• ^[1] (Chosun Ilbo) – Paragraph 1, Paragraph 3, Paragraph 6
• ^[2] (AP News) – Paragraph 1, Paragraph 2, Paragraph 4
• ^[3] (Tom’s Hardware) – Paragraph 3
• ^[4] (Tom’s Hardware) – Paragraph 2, Paragraph 6
• ^[5] (Tom’s Hardware/Bernstein) – Paragraph 5
• ^[6] (Axios) – Paragraph 4
• ^[7] (Forbes) – Paragraph 4

Source: Noah Wire Services