U.S. Chip Rules Redraw AI Supply Lines With New Export Tests and Tariffs

WASHINGTON — A U.S. Commerce Department rule that took effect on Jan. 15 changed the review standard for some advanced AI chip exports to China while imposing new certification requirements, and a related presidential proclamation added a 25% tariff on covered advanced chip imports intended for customers outside the United States, according to a January legal analysis by Mayer Brown.

The rule applies to chips performing at or below the level of Nvidia's H200 and AMD's MI325X, the law firm said in its summary of the Bureau of Industry and Security action. It said the review posture moved from a general presumption of denial toward case-by-case review for certain transactions, while expanding the information exporters must provide.

The same analysis said applicants would have to certify technical, business and end-user conditions, including security procedures, identification of remote end users in countries of concern and evidence that exports would not divert foundry capacity from U.S. buyers.

The administration presented the move as a more flexible policy that still protected national security. Congress, according to the same analysis, continued to press concerns about advanced AI capabilities reaching China.

That tension is now shaping how the AI supply chain is described in different capitals. In Washington, the issue is framed as a national security filter meant to allow lower-risk commercial flows while restricting military or sensitive end uses. In Seoul and Taipei, the same measures are part of a harder question about what product mix, fab capacity and customer planning look like when policy can alter the market mid-cycle.

The scans reviewed by Albis on April 8 showed that Korean coverage had been treating AI hardware less as an abstract race and more as a compute access problem touching data centers, cloud costs and the reliability of expansion plans. That is a different emphasis from U.S. political debate, where hearings and export control language dominate.

The trade effects are already showing up in broader data. UNCTAD said this week that strong demand for AI-related goods and digital technologies had been one of the main engines of manufacturing trade growth in 2025, even as protectionism and shipping disruption clouded the wider outlook.

That means AI chips now sit at the intersection of two opposing trends. On one side, demand remains strong enough to support trade growth. On the other, licensing tests, tariffs and geopolitical restrictions are making supply more conditional.

Mayer Brown said the tariff applied to covered advanced AI chips that were not destined for the U.S. technology supply chain. The firm also said the rule required third-party testing in the United States to verify performance specifications before export.

Those details may sound technical, but they change timelines and bargaining power. Exporters now face more documentation, more screening and more uncertainty over which customers and destinations will clear review.

The policy also reaches beyond direct U.S.-China trade. Mayer Brown said applications involving third-country remote end users would face heightened disclosure obligations and scrutiny, which means allied and partner markets can be pulled into the same compliance web.

That broader spillover is one reason regional framing differs. For U.S. officials, third-country routing is part of enforcement. For governments in Southeast Asia and the Gulf, it can look like a warning that even indirect participation in the AI hardware trade may face additional checks.

The January changes did not shut down all advanced chip sales, and the administration described them as more targeted than earlier blanket restrictions. But they reinforced the idea that compute has become a managed strategic resource rather than a standard technology export.

The next test will come as exporters file applications under the revised standard and as manufacturers decide where to allocate limited advanced capacity. If approvals slow or tariffs bite harder than expected, the result will not stay inside the chip sector. It will feed into cloud pricing, data-center buildouts and the geography of AI deployment through the rest of 2026.