The 25% Chip Tariff Is Live. Here's Who Pays, Who Doesn't, and What to Do This Week.

If you're importing NVIDIA H200s or AMD MI325X chips for a project outside the US, you just got 25% more expensive. On January 15, 2026, a new tariff took effect targeting the most advanced AI chips on the market.

But here's what the headlines miss: most US-based buyers are exempt. The tariff is designed to penalize foreign AI infrastructure while protecting domestic investment. If you're building in the US, you're probably fine. If you're supplying chips to data centers in Europe, India, or anywhere else, you're paying the premium.

The distinction matters because the exemption process isn't automatic. You need the right documentation, the right customs classification, and a clear paper trail showing your chips are headed for qualifying US uses. Get it wrong, and Customs will assess the full 25%.

Here's what you need to know.

Why AI Data Centers Are Different (And Why That Matters Here)

Not all data centers are created equal, and this tariff draws a sharp line based on what's inside them.

Traditional data centers run general computing workloads: web servers, databases, email systems. They use standard processors and have relatively modest power requirements. An AI-focused data center is a different animal. Training large language models and running AI inference at scale requires specialized chips that can handle massive parallel calculations. These chips, like the NVIDIA H200 and AMD MI325X targeted by this tariff, consume far more power and generate far more heat than traditional server hardware.

That's why the tariff exemption uses a 100 megawatt threshold. A typical enterprise data center might need 10-30 MW. A hyperscale AI training facility can require 500 MW or more. The 100 MW cutoff is designed to capture large-scale AI operations while excluding smaller server deployments.

Why does this matter for tariffs? The US government wants AI development to happen on American soil. By making these chips more expensive for buyers building AI infrastructure overseas, the tariff creates a financial incentive to build data centers in the United States instead.

What the Administration Did

On January 14, 2026, President Trump signed a Presidential Proclamation imposing a 25% tariff on certain advanced computing chips. The legal authority comes from Section 232 of the Trade Expansion Act of 1962, which allows the President to restrict imports that threaten national security. The tariff took effect at 12:01 AM EST on January 15, 2026.

What "25% ad valorem" means: This is trade jargon for "25% of the value." If you import a chip worth $10,000, you pay $2,500 in tariff duties on top of the purchase price.

Unlike the broader semiconductor tariffs the administration has signaled, this one is narrowly targeted. It covers only the most advanced AI chips, specifically those meeting certain technical performance thresholds. In practical terms, this means chips like the NVIDIA H200 and AMD MI325X.

The tariff applies to imports classified under specific customs codes (called HTS codes): 8471.50, 8471.80, and 8473.30. US Customs and Border Protection (CBP) has issued guidance (CSMS #67400472) with detailed instructions on how to classify these products.

Who's Exempt (and Who Isn't)

The proclamation carves out seven categories of uses that don't have to pay the tariff. Each exemption has its own customs code that importers must use when filing paperwork:

Exempt (0% duty):

• US data centers: If you're building or operating a data center in the United States that requires more than 100 megawatts of power for AI workloads, you're exempt. For context, 100 megawatts can power roughly 80,000 homes. This threshold targets large-scale AI facilities, not small server rooms.
• Repairs and replacements: If you're bringing chips into the US to fix or replace existing equipment, you're exempt.
• Research and development: Academic institutions, corporate R&D labs, and others conducting research in the US are exempt.
• Startups: Companies that qualify as "emerging growth companies" under securities law are exempt. This generally means companies with less than $1.235 billion in annual revenue that have been public for fewer than five years, or are still privately held.
• Consumer electronics: Gaming systems, personal computers, and automotive applications are exempt. If you're building PlayStations or laptops, not data centers, you don't pay the tariff.
• Industrial applications: Robotics, manufacturing equipment, and other industrial uses outside of data centers are exempt.
• Government and defense: Public sector applications are exempt.

The Secretary of Commerce can also grant exemptions for other uses that "strengthen the US technology supply chain or domestic manufacturing capacity."

Who pays the tariff:

• Companies importing chips to build data centers outside the US
• Resellers shipping to foreign buyers
• Anyone building AI infrastructure in Europe, Asia, or elsewhere

The bottom line: If you're Amazon Web Services building a data center in Virginia, you're exempt. If you're a cloud provider building in Frankfurt, Germany, you're paying 25% extra.

Three Scenarios to Understand How This Works

The rules can get confusing, so here's how the tariff plays out in three common situations:

Scenario 1: US Company Importing Chips for a US Data Center

The setup: A US cloud provider orders 1,000 NVIDIA H200 chips from a distributor. The chips are manufactured in Taiwan, shipped to the US, and installed in a data center in Texas that draws 150 MW for AI workloads.

What happens: This is the straightforward case. The company files customs paperwork using the data center exemption code (9903.79.03), provides documentation showing the chips are for a qualifying US facility (lease agreements, power contracts, construction permits), and pays 0% tariff. The chips enter the country duty-free.

Key requirement: The data center must require more than 100 MW of new load for AI workloads. If it's a smaller facility, the company would need to qualify under a different exemption (startup, R&D, etc.) or pay the 25%.

Scenario 2: US Company Building a Data Center in Germany, Shipping Directly

The setup: A US tech company is building a data center in Frankfurt. They order H200 chips and have them shipped directly from Taiwan to Germany. The chips never touch US soil.

What happens: The US tariff does not apply. US Customs can only impose duties on goods that enter US territory. If the chips ship directly from Taiwan to Germany, they bypass US customs entirely.

However, this doesn't mean it's a clean workaround. The chips are still subject to US export control rules because they contain US-origin technology. Even though they're manufactured in Taiwan, NVIDIA and AMD must comply with US export licensing requirements. The company may face restrictions on which countries can receive the chips, volume limitations, and end-use requirements, but those are export controls, not tariffs.

The practical impact: No 25% US tariff, but the company will pay whatever import duties Germany and the EU impose, and must still comply with US export regulations.

Scenario 3: A Chinese Company Wants to Buy These Chips

The setup: A Chinese AI company wants to purchase H200 chips for a data center in Shanghai.

What happens: This is where the tariff and export controls intersect, and it gets complicated.

First, the chips cannot ship directly from Taiwan to China. Under the proclamation, chips destined for China must route through the United States for independent third-party testing to verify their AI capabilities.

Why the US routing requirement exists: The US government has limited visibility into what happens to chips once they leave Taiwan and arrive in China. By requiring China-bound chips to physically pass through the United States first, the government creates an enforcement checkpoint. Third-party testing verifies that the chips match their declared specifications and haven't been modified. It also creates a paper trail documenting exactly who is buying what, in what quantities, and for what stated purpose.

This matters because of diversion risk. A chip sold for a "commercial cloud computing" application could theoretically be redirected to military AI systems, surveillance infrastructure, or other uses the US government wants to restrict. When chips ship directly from Taiwan to China, there's no practical way for US authorities to verify end use. Routing through the US forces the transaction into a controlled environment where inspections, documentation, and enforcement actions are possible.

In short: the routing requirement turns a paperwork exercise into a physical checkpoint.

Second, the Chinese buyer pays the full 25% tariff. None of the exemptions apply because the chips aren't being used in the US.

Third, there's a volume cap. Shipments to China are limited to 50% of the seller's US sales volume. If NVIDIA sells 10,000 H200 chips to US customers, they can only sell 5,000 to Chinese buyers.

Fourth, existing export controls still apply. Depending on the specific chip model and the Chinese buyer's identity, additional Commerce Department licenses may be required, or the sale may be prohibited entirely if the buyer is on the Entity List.

The practical impact: Chinese buyers face the 25% tariff, mandatory US routing, third-party testing, volume caps, and potentially additional export licensing requirements. This is intentionally burdensome.

What This Actually Means for Your Business

The tariff creates a two-tier pricing structure for the world's most advanced AI chips. US-based AI projects just got relatively cheaper compared to the rest of the world. That's exactly what the policy is designed to do.

If you're a large tech company building AI infrastructure in the US: Your costs haven't changed. File under the data center exemption with supporting documentation proving your US location and power requirements.

If you're a startup: The emerging growth company exemption likely applies to you. You'll need to show that you meet the legal definition, which your lawyers or accountants can help verify.

If you're importing for customers outside the US: You're the target of this tariff. Plan for 25% on top of your current costs, or restructure your supply chain so that foreign customers import the chips directly themselves.

If you're a contract manufacturer: This gets complicated. If you're assembling products in the US with these chips for US customers, you're likely exempt. If you're assembling for export, you're not. The exemption follows where the chip ends up being used, not where the importer is located.

What You Need to Do to Claim an Exemption

The exemptions don't happen automatically. You need to actively claim them with proper documentation. Here's what that involves:

1. Use the correct customs code: When you file import paperwork, you must use the specific exemption code that matches your situation, not the general code that triggers the 25% duty.

2. Prove your exempt use: Customs expects documentation showing why you qualify. This could be data center construction permits, lease agreements, R&D project descriptions, or proof of your startup status.

3. Keep exempt and non-exempt inventory separate: If you're importing chips for both exempt uses (like your US data center) and non-exempt uses (like resale overseas), you need tracking systems to prove which chips went where. Mixing inventory creates audit problems.

4. Understand Foreign Trade Zone rules: Foreign Trade Zones are special customs areas where companies can defer paying duties. For these chips, the usual deferral rules don't apply. You'll pay the duty when the chips enter US commerce.

5. Know that refunds are limited: If you pay the 25% and later realize you should have been exempt, you can't get the money back through normal refund programs. You'll need to file an administrative claim, which is slower and less certain.

For questions, CBP's contact is TradeRemedy@cbp.dhs.gov. Expect delays; they're getting many inquiries.

How This Affects Contracts

If you're buying or selling these chips, your contracts probably need updating.

Purchase agreements: Clarify who bears the tariff risk. Standard trade terms like "DDP" (delivered duty paid) mean the seller covers all duties, which now includes a potential 25% hit. "DAP" (delivered at place) pushes that cost to the buyer. Don't assume the other party is handling classification correctly.

Reseller agreements: If you distribute these chips, make sure your contracts address who certifies the end use. You don't want to claim an exemption as the importer, only to discover your customer is reselling the chips overseas.

Data center construction contracts: If you're a large tech company hiring contractors to build your data center, verify that hardware procurement flows through exemption-eligible channels. A contractor who imports chips without proper paperwork exposes everyone to duty liability.

Investment deals: For AI startups raising capital, expect questions from investors about tariff exposure. If your business model involves building infrastructure outside the US, your cost structure just changed significantly.

The Taiwan Carve-Out

There's a special deal for companies sourcing chips from Taiwan, where most advanced semiconductors are manufactured (including at TSMC, the world's largest chip foundry).

The administration announced a parallel agreement rewarding Taiwanese chip makers that invest in US manufacturing. Companies building new chip factories in the US can import up to 2.5 times their planned US production capacity duty-free during the construction period. Imports above that quota get a reduced rate.

This is designed to accelerate construction of chip factories on US soil. If you're working with a Taiwanese supplier that's building US capacity, ask about their quota allocation.

What Comes Next

This tariff is a down payment on something bigger. The administration has consistently signaled broader semiconductor tariffs are coming. Here's what to watch:

Practical Takeaways

Here's your compliance checklist for this week:

• Audit your current imports: Identify any shipments of NVIDIA H200, AMD MI325X, or comparable chips. Verify they're being classified under the correct exemption code.
• Document your exempt use: Gather evidence supporting your exemption claim. This includes data center leases, construction permits, R&D project descriptions, or proof of emerging growth company status.
• Update your customs procedures: Work with your customs broker to ensure import filings after January 15 use the new codes. Mistakes are hard to fix after the fact.
• Separate your inventory: If you import for both exempt and non-exempt purposes, set up tracking systems now. Mixed inventory invites audits.
• Review supplier contracts: Confirm who bears tariff risk under your purchase terms. Renegotiate if the current terms don't account for this change.
• Brief your finance team: If you're paying the 25%, update your cost projections. If you're claiming exemptions, budget for the compliance overhead.
• Mark July 1 on your calendar: The Commerce Department's report could change the rules mid-year.

Looking Ahead

The 25% tariff on advanced AI chips is targeted, exemption-heavy, and clearly designed to tilt the economics of AI development toward the US. For most domestic buyers, it's mainly a paperwork exercise. For companies building AI infrastructure globally, it's a real cost increase.

The compliance burden is manageable if you're prepared. The bigger question is what comes next. If this tariff becomes the template for broader semiconductor restrictions, the current generous exemption structure may not last. Build your documentation systems now, while the rules are clear.

This analysis reflects publicly available information as of February 2, 2026. Companies should consult qualified trade counsel for compliance guidance specific to their operations.