Trump imposes 25% AI chip tariff as supply priority shifts to U.S. buyers

“Artificial intelligence is not just a software story,” wrote the New York Times. “It’s a routing story.” A chip’s path, where it’s made, where it’s tested, and where it clears customs, can change the final price as much as the performance. That is the subtle lever driving the White House’s current action on advanced AI hardware.

On January 15, 2026, President Donald Trump imposed an immediate 25% tariff on a specific category of “advanced computing chips” under a Section 232 national security order. According to a White House factsheet, Nvidia’s H200 and AMD’s MI325X are among the products covered following a nine-month Section 232 investigation into supply chain vulnerabilities.

The detour that turns U.S. customs into a toll point

The headline rate is 25%, but the mechanism is the part most people will miss. Reuters reports the administration now requires that China-bound semiconductors routed through the U.S. are subject to the 25% duty, making U.S. customs a collection point even when the chips are not staying in the United States.

This intersects with a Commerce Department policy allowing limited H200 exports to China, provided they meet strict conditions, including third-party testing in U.S. labs to verify technical qualifications. Commerce’s conditions also add three constraints that tighten the channel further:

A 50% cap: China-bound shipments may not exceed 50% of the volume delivered to U.S. customers.

Blackwell and Rubin left out: The carveout does not include Nvidia’s new Blackwell line or the next-generation Rubin platform.

Supply priority: Nvidia must prioritize fulfilling U.S. domestic demand before China shipments proceed.

Tariff triage: who pays and who is shielded

The White House frames this as a targeted action rather than a broad semiconductor tariff. The structure indicates that domestic AI development is intended to remain protected.

Nvidia H200

Tariff logic: A 25% Section 232 duty applies to “advanced computing chips” in instances where H200-class hardware enters U.S. customs as part of U.S.-routed exports.

Who is exempt: Importers related to U.S.-based data centers, AI startups, R&D, and other domestic-use categories.

AMD MI325X

Tariff logic: The MI325X is cited as an example of the same “advanced computing chips” category, where coverage is defined by performance standards rather than brand names.

Who’s exempt: The same system of exemptions applies to key domestic uses, including consumer, civil industrial, and public uses that facilitate the U.S. buildout.

Derivative products

Tariff logic: Certain derivative products can be pulled into scope based on how the annex defines them and how enforcement interprets “derivative” scope.

Who’s exempt: The carveouts aim to prevent hardware from being taxed in cases where the import contributes to building U.S. supply-chain capacity.

The “Covered” Determination

The government isn't writing this policy around brand names; it’s writing it around capability. In the relevant export control regime, eligibility is based on Total Processing Performance (TPP) and memory bandwidth. Current criteria for the new China licensing approach include TPP levels under 21,000 and DRAM bandwidth below 6,500 GB/s.

A useful way to read the policy stack is this: the definition of “advanced computing” is beginning to function like a tax bracket. If a chip is powerful enough to trigger licensing scrutiny for export, it is effectively in the category Washington wants to tax at the border.

Why Taiwan and TSMC sit at the center

As per Reuters, the Section 232 move is directly tied to worries about dependence on foreign sources, especially Taiwan. That’s the state of manufacturing for most of the AI stack: the accelerators are designed by U.S. firms, but the leading edge of manufacturing happens in Taiwan’s ecosystem, with TSMC as the pivotal player.

Who Controls Exemptions

The “gate” within this policy is located in Commerce. Reuters reported that exemption authority within this structure rests with Commerce Secretary Howard Lutnick, giving the department practical leverage over edge cases where eligibility could swing hardware costs.

Market Reaction: A Toll, Not a Crash

The shares of Nvidia, AMD, and Qualcomm fell slightly in the wake of the tariff announcement. One reason the reaction remained contained is that the tariff explicitly exempts chips intended for use inside U.S. data centers, which is the heart of global demand.

Key Takeaway: The policy prioritizes U.S. AI infrastructure by shielding data centers and startups while imposing an export supply-priority requirement. China-linked flows face layered friction through U.S.-based third-party testing, a 50% volume cap, and continued limits on Blackwell/Rubin-class chips. 

For investors, the initial response suggests this is a "policy toll" rather than a demand shock, though the real swing factor will be whether the definition of "covered products" expands.