Administration Policies on Advanced AI Chips Codified, with Reverberations Across AI Ecosystem

Commerce rule eases licensing for advanced AI chips to China and establishes new criteria for license applications; Presidential Proclamation imposes 25% tariff on covered advanced chip imports when intended for any customer outside United States

In the last week, the Trump Administration has taken several actions to formalize policies enabling the export of advanced AI semiconductor chips to China and PRC companies. Each measure has repercussions across the AI ecosystem.

On January 13, 2026, the US Department of Commerce’s Bureau of Industry and Security (“BIS”) released a final rule revising the license review posture for commercially available NVIDIA H200- and AMD MI325X-equivalent chips and chips with lesser performance from “presumption of denial” to “case-by-case review.” With this rule, BIS established specific technical, business, end user, and US market certifications exporters must make to benefit from the revised policy, including identification of their customers’ remote end users in a specific list of countries of concern.  

The following day, hours after the House Foreign Affairs Committee completed its hearing on “Winning the AI Race Against the Chinese Communist Party,” the President released a Proclamation imposing a 25 percent tariff on the import of advanced AI chips. The tariff applies to all covered advanced AI chips that are not for the US technology supply chain.

Bottom Line Up Front

US policy on advanced AI to China remains in flux. As the Administration codifies licensing policy and national security-based tariffs, Congress continues to express concern about the export of advanced AI capabilities to China. China’s receptivity to US advanced AI imports remains uncertain. In the meantime, US government measures to revise China policy will reverberate across the AI ecosystem, affecting all types of entities and sales to ally and partner countries.

Advanced AI Chips Rulemaking

BIS’s final rule, which became effective on January 15, 2026, formalizes a more flexible license review policy for transactions involving H200- and MI325X-equivalent chips and lesser-performing chips. The policy prioritizes national security objectives through transaction-specific risk assessments—particularly with respect to sensitive end uses and end users—while avoiding sweeping prohibitions that could restrict lower-risk activities. At the same time, BIS expands the information license applicants must submit.

In substance, the rule essentially codifies statements previously made by the Trump Administration on social media in December to President Xi Jinping of China that the United States will allow NVIDIA to ship its H200 products to China and other countries “under conditions that allow for continued strong national security.”

Applicants must also certify that their customers demonstrate “sufficient” security procedures. Applicants must also certify that there is “sufficient” supply of the product in the United States and manufacture of the product for China export purposes will not divert foundry capacity from US end users. In addition, the product must undergo third-party testing in the United States to verify performance specifications before export, with applicants certifying this step to the US government.

Rule Scope

• Product scope: The rule applies to chips that perform at or below the level of the H200 and MI325X based on performance and technical characteristics referenced in recent advanced computing controls.
• Geographic scope: China remains the principal national security focus, but BIS directs attention to third-country remote end users. Applications involving such destinations will face heightened disclosure obligations and scrutiny.
• End-use and end-user risk: Transactions linked to military, intelligence, surveillance, or other sensitive programs—or involving restricted parties—remain subject to rigorous review and likely denial, irrespective of destination.

New License Application Requirements

As with all export licensing, applicants must certify and provide necessary supporting data as part of the licensing process as part of the new review policy.

For exports to China and Macau under the new rule, applications must provide a host of new information to BIS, only some of which has been historically included in license applications. 

From a technical perspective, applications must certify that the items operate below the performance criteria, and provide total processing performance (“TPP”), total DRAM bandwidth, interconnect bandwidth, copackaged DRAM capacity, peak power at max TPP, as well as explanations for any changes to specifications. They also must certify that the AI chips will undergo review by “a qualified third-party testing lab” to confirm these technical capabilities and functions.

With respect to the US market, applicants must:

• Identify the number of units of the item at issue that have been shipped in the United States;
• Certify with evidentiary support that the aggregate TPP exported to China and Macau is less than 50 percent of aggregate TPP shipped to US customers for US end use; and
• Certify that there is sufficient supply of the product in the United States, that such export would not result in any delay for customers in the United States for end use in the United States, and that exports will not divert global foundry capacity otherwise used for similar or more advanced chips for US end users.

Finally, the rule adds several requirements with respect to customer due diligence, continuing the past year’s trend of requiring exporters to engage in more substantial Know Your Customer (KYC) activity. Applicants must:

• Confirm that the transaction is not prohibited by end user/use controls and controls for nonmilitary end uses/end users, and that no prohibited parties will be granted remote access;
• Obtain and submit the KYC procedures of their ultimate customer to prevent unauthorized remote access to unauthorized parties;
• Identify their customers’ end users in named third-country jurisdictions (Belarus, China, Cuba, Iran, Macau, North Korea, Russia, and Venezuela), including legal names, addresses, ownership/affiliations, intermediaries, and ultimate end users, to address transshipment risks;
• When the ultimate customer or end user provides AI as a Service, verify additional commitments, including no undisclosed transfer of model weights and no prohibited party remote access to algorithms trained on the chips; and
• Describe the physical security measures of the ultimate consignee.

BIS continues its recent trend of expecting exporters to increase due diligence—in this case, on customers and customers’ end users when AI as a Service is provided, as well as through third-party review of technical performance.

Enforcement Efforts

BIS is expected to expand the scope of its monitoring and confirmation of license application assertions consistent with its FY26 budget increase through which the agency is set to receive at least $10 million targeted for “enforcement of the restrictions on end uses and end users for products classified under Export Control Classification Number 3A090 and related technologies.” The FY26 appropriations package that funds BIS has passed both Chambers in Congress and is destined to be enacted into law ahead of the January 30 deadline. While the President has full authority to veto, we do not anticipate such event occurring.

Who is Affected

The government’s signals about end users in countries of concern, security measures, foundry capacity, and the US market should be noted by all AI industry actors and not just exporters of the targeted chips. These measures are particularly relevant to:

• Semiconductor designers and manufacturers shipping H200-class or equivalent chips;
• OEMs and integrators incorporating controlled chips into servers or clusters;
• Cloud service providers and colocation data centers offering access to controlled compute;
• Distributors and channel partners with activities in specified third countries; and
• Multinational enterprises deploying controlled chips for AI training or inference across borders.

Section 232 Tariffs on Certain Covered AI Chip Imports

On January 14, 2026, President Donald Trump issued a Proclamation, Adjusting Imports of Semiconductors, Semiconductor Manufacturing Equipment, and Their Derivative Products into the United States. Effective January 15, the President imposed a 25 percent value-based tariff on “covered products.” Covered products are listed in an Annex to the Proclamation and, as a practical matter, in-scope covered products are advanced AI chips that are not destined for the US supply chain. Thus, while the tariff is not specific to China, it essentially implements President Trump’s December social media post noting that 25 percent of H200 sales to China would be “paid to” the United States.

The in-scope covered products are logic integrated circuits and items containing a logic integrated circuit with either: (1) TPP between 14,000 and 17,500 and a total DRAM bandwidth between 4,500 and 5,000 GB/s; or (2) TPP between 20,800 and 21,100 and total DRAM bandwidth between 5,800 and 6,200 GB/s.

Excluded from the tariff (by designation under specific eight-digit tariff codes) are products imported for the following purposes:

• Use in US data centers;
• Repairs or replacements performed in the United States;
• R&D development in the United States;
• Use by startups;
• Non-data center consumer applications in the United States;
• Use in non-data center civil industrial applications in the United States; or
• Use in US public sector applications.

Notably, the Proclamation Annex defines a US data center as a facility with more than 100 MW of new load dedicated to AI inference, training, simulation, or synthetic data generation.

The Proclamation leaves open the possibility that other exclusions may be identified by the Secretary of Commerce based on a determination that the imported items contribute to strengthening the US technology supply chain or domestic US manufacturing capacity.

Entities that bring covered chips into the United States before selling to foreign customers will be affected by 25 percent tariffs, even if they are not exporting to China.

Congressional Approach

How forceful the US Congress will be on permissive exports of advanced AI chips to China remains an open question.

During the January 14, 2026, House Foreign Affairs Committee (“HFAC”) hearing on “Winning the AI Race Against the Chinese Communist Party,” Chairman Brian Mast (R-FL) noted that private companies in China purchase chips “for applications that will ultimately assist militaries of our foreign adversaries” and that PRC private companies are “working to make their military more dominant by buying our chips.” Experts and members of Congress overwhelmingly expressed skepticism regarding the Administration’s approach to exports of advanced AI chips to China. As Trump Administration first term Deputy National Security Advisor Matt Pottinger put it, “Congress needs to step in, reverse the policy, and put durable guardrails in place so the mistake can’t be repeated.”

Along these lines, in December 2025, Chairman Mast introduced the AI Overwatch Act, which would require congressional review of export licenses of advanced AI chips to China. During the January 14 hearing, Mast stated, “Should Congress have oversight when selling missiles to other countries? Yes, the same should be said for chips.” This bill covers all items designated under Export Control Classification Numbers 3A090, 4A090, and functional equivalents, and would require a 30-day notification period to Congress before a license may be issued for exports to China, Cuba, Iran, Macau, North Korea, Russia, and Venezuela (similar to such requirements for military sales, and more stringent than BIS’s existing Missile Technology certification requirement under Section 1512 of the FY1999 National Defense Authorization Act).

Similarly, other pending legislation would also enhance Congress’ role in overseeing the Administration’s China AI policy. These include the:

• Guaranteeing Access and Innovation for National Artificial Intelligence (GAIN AI) Act of 2025 (would require companies to give US businesses first priority in acquiring advanced AI chips before exporting these chips to China and other countries of concern);
• Semiconductor Technology Resilience, Integrity, and Defense Enhancement (STRIDE) Act (would require the State Department to coordinate with partner countries to enhance their semiconductor technology export controls);
• China AI Power Report Act (would require the Commerce Department to submit annual reports to Congress on China’s advanced AI capabilities); and
• Divesting from Communist China’s Military Act (would require that PRC military companies listed on the Pentagon’s 1260H List be also listed on the Department of the Treasury’s list of Non-Specially Designated National Chinese Military Industrial Complex list).

Also in this vein, on January 12, the House of Representatives passed the Remote Access Security Act, which would expand the scope of US export controls to cover the provision of cloud-based services. BIS’s regulatory requirement that license applicants provide details on their PRC customers’ remote end users appears to align with this effort.

China’s Response

It remains to be seen if China will actually buy the AI chips covered in these measures. Recent reporting indicates that PRC Customs has placed a temporary halt on clearing H200 chips to enter China. PRC government officials also reportedly summoned domestic companies to meetings in which the companies were explicitly instructed to not purchase the chips unless necessary.

Conclusion

The export control and tariff measures formalize and demonstrate continuity in the Administration’s approach to providing advanced AI to China. They also provide new clarity on the conditions under which exports may be approved. Both actions appear targeted to the business practices of major chip designers and focused on China, suggesting that clarifications may be coming, including for third-party resellers and for companies that import advanced AI chips before selling to allied and partner country markets.