Here’s what you actually need to know: the 25% AI chip tariff under Presidential Proclamation 11002 is not a chip ban. Chips can still legally move to China — they just cost more. For most companies buying GPUs for their own data centres, dev environments, or cloud deployments, the tariff is essentially a non-event. A broad set of use-based exemptions means domestic procurement is almost entirely untouched.
What we’re going to cover here: how the tariff defines which chips are in scope, why Section 232 was chosen as the legal vehicle, how the transshipment trigger actually works, what the BIS approval process looks like, and how this compares to what Biden was doing. The April 14, 2026 Phase 2 deadline has passed at time of writing — what comes next is still unconfirmed.
For the broader picture of how this fits into hardware budget planning, our comprehensive AI chip tariffs and semiconductor geopolitics guide for 2026 covers cost scenarios end-to-end.
What Is the 25 Percent AI Chip Tariff, and Which Chips Does It Actually Cover?
Proclamation 11002, signed January 14, 2026, puts a 25% ad valorem duty on advanced computing chips that don’t qualify for a use-based exemption. It applies at the point of importation into the United States.
Coverage is defined by technical specifications, not brand names. A chip has to clear a Total Processing Performance (TPP) floor and a DRAM bandwidth floor at the same time — both thresholds must be met simultaneously.
The Nvidia H200 and AMD MI325X are in scope. The H100 and A100 fall below the thresholds. Blackwell (B200) is categorically off-limits for China via Export Administration Regulations — a separate mechanism we’ll get to below.
Covered chips use HTSUS classification 9903.79.01. Exempt imports use 9903.79.03 through 9903.79.09 depending on the applicable use category. Most US buyers never encounter the tariff at all. Data centre, R&D, startup, consumer, industrial, and public sector uses all qualify for exemptions. The tariff is targeted at chips moving through the US supply chain on their way to China — not at domestic buyers.
What Is Section 232, and Why Does the Legal Basis of This Tariff Matter?
Section 232 of the Trade Expansion Act of 1962 (19 U.S.C. 1862) gives the President the authority to impose import restrictions on products whose importation threatens national security — after a formal Commerce Department investigation. No market access violation needs to be proven. National security is the only threshold.
Commerce kicked off the semiconductor investigation in April 2025 and transmitted the final report on December 22, 2025. Proclamation 11002 followed in January. The finding: the US fully manufactures only around 10% of the chips it needs.
So why Section 232 rather than emergency-powers IEEPA? Legal durability. The steel and aluminium Section 232 tariffs from 2018 survived court challenge — this tariff sits on the same statute. IEEPA-based tariffs face a higher constitutional challenge risk. If you’re planning hardware budgets, legal stability is something you can actually model against.
There’s also a cost-clarity benefit worth knowing about. Chips subject to Section 232 are explicitly excluded from IEEPA tariff stacking — no compounding duty from both regimes. The exception: the 40% IEEPA tariff on Brazil and the 10% IEEPA tariff on China do stack for chips from those origins.
Trump’s own framing makes the intent pretty plain: “We’re going to be making 25 percent on the sale of those chips, basically. So we’re allowing them to do it, but the United States is getting 25 percent.” Revenue generation alongside deterrence — not a prohibition.
For more on how legislative developments could affect this framework, the AI Overwatch Act and Remote Access Security Act covers the legislative risk.
What Is Transshipment, and Why Is It the Tariff’s Triggering Mechanism?
Transshipment is the key concept to get your head around. It’s the process of importing chips into the United States and then re-exporting them to China or Macau — using US territory as a pass-through.
The flow looks like this: TSMC manufactures chips in Taiwan → chips ship to the US → importer processes or tests them → chips re-export to a Chinese buyer. That re-export pathway triggers the 25% tariff under HTSUS 9903.79.01.
Chips arriving for genuine US domestic use qualify for use-based exemptions and pay zero additional tariff. Any semiconductor eligible to ship to China must first be imported into the United States for third-party performance testing, at which point it becomes subject to the 25% tariff — that’s how the US collects its cut of H200 sales heading to China.
If you’re procuring H200s for your own data centre or dev environment, you’re almost certainly exempt. The indirect risk is that if your supplier transshipped chips without an exemption, that cost can show up in your unit price.
BIS determines whether an export is legally permissible; Section 232 determines what duty applies if it is. Both must be satisfied — they’re addressing different aspects of the same transaction.
For more geopolitical context, the US-Taiwan $500 billion semiconductor deal and its implications for AI chip access goes into this in more detail.
What Is the BIS Approval Process, and What Criteria Must an H200 Export to China Satisfy?
The Bureau of Industry and Security (BIS) runs export controls under the Export Administration Regulations (EAR). It operates separately from the Section 232 tariff but applies in parallel to any China-bound chip export.
On January 15, 2026, BIS published a final rule shifting export licence review for H200 and MI325X to China from “presumption of denial” to case-by-case review — the first relaxation of a three-year denial policy. That’s not automatic approval. Each application is assessed on its own merits.
There are four criteria an applicant has to satisfy:
US Supply Protection — the export won’t reduce semiconductor production capacity available to US customers.
Know Your Customer — the Chinese purchaser has adopted export compliance procedures including customer screening.
Third-Party Testing — products undergo independent testing in the United States to verify performance and security.
Foundry Capacity — production won’t divert foundry capacity away from US-priority customers.
There’s also a volume cap: aggregate shipments to China and Macau can’t exceed 50% of total product shipped to US customers — an independent ceiling regardless of individual approvals. And Chinese authorities retain final import approval — US approval doesn’t guarantee access at the China end.
One hard rule: Nvidia Blackwell is categorically off-limits. No BIS licence pathway exists.
For more on how the BIS approval criteria interact with Chinese procurement strategy, why Chinese tech companies ordered 2 million H200s despite regulatory risk covers the demand-side mechanics in depth.
How Does the Trump Tariff Differ from Biden-Era Chip Export Controls?
Biden’s approach (2022–early 2025): outright prohibition on exports of H100, A100, and H200 to China via EAR/BIS rules. No price mechanism. A legal bar on the transaction, full stop.
Trump’s approach, in effect from January 2026: replace the blanket prohibition with a tariff. H200s can legally reach China at a 25% premium. Biden’s EAR controls blocked access regardless of price. Trump’s tariff allows access at a cost premium you can actually model. You can build 25% into a procurement scenario in a way you simply can’t do with a binary prohibition that might change overnight.
The H20 episode makes this concrete. Trump initially banned the H20 in April 2025, then reversed course by August 2025 in exchange for Chinese relaxation of critical minerals controls. H20 exports are currently permitted. “Price not prohibition” playing out in practice.
For a full analysis of how this translates to budget impact, the full semiconductor geopolitics overview walks through the cost scenarios in detail.
Which Chips Are Still Completely Off-Limits for China, and Where Does the H200 Sit?
Here is the GPU hierarchy for China-bound exports as of April 2026:
Blackwell B200: categorically off-limits under EAR. No licence pathway. Nvidia’s Vera Rubin architecture will almost certainly face the same treatment.
Nvidia H200 / AMD MI325X: licensable via BIS case-by-case review, subject to the four criteria and the 50% volume cap. 25% tariff applies on US-transshipped supply. The highest-performance chips currently accessible to Chinese buyers — the top of the permitted range.
Nvidia H100 / A100: below the Section 232 thresholds. Tariff doesn’t apply. Separate BIS/EAR rules still restrict China-bound exports.
Nvidia H20: currently permitted, below threshold. No tariff applies. Deliberately performance-limited China-specific Hopper variant.
Why does China want the H200 over the H20? The H20 was engineered to fall below export control thresholds — TPP and DRAM bandwidth are cut significantly. That performance gap matters for training large models. CUDA lock-in compounds it: switching to Huawei Ascend requires rebuilding the ML stack from scratch — and Huawei’s Ascend 910C is roughly 60% worse than the H100 in real-world inference performance, a chip from 2019.
For Western buyers: entities bringing covered chips into the US before selling to foreign customers face 25% tariffs even if not exporting to China — the tariff triggers at US importation.
For more on how HBM3e supply constraints interact with tariff costs, HBM4 delays and GDDR7 shortages squeezing GPU supply in 2026 covers that dimension.
What Does Any of This Mean for Companies Buying AI Hardware in 2026?
The short answer: for most companies buying GPUs for their own data centres, cloud deployments, or development environments, the tariff has essentially zero direct cost impact. The Proclamation explicitly exempts US data centres, R&D, startups and emerging growth companies, consumer and industrial applications, and public sector procurement.
There is one gap worth keeping an eye on. “Startups” and “emerging growth companies” are listed as exempt, but no formal definition or size threshold has been published by CBP. If your company might plausibly qualify, keep contemporaneous records and watch for CBP guidance.
For companies re-exporting or redistributing chips — resellers, system integrators supplying Chinese customers — the 25% tariff is a direct cost you need to account for.
On the admin side: importers must confirm which exemptions apply and notate the appropriate Chapter 99 HTSUS code on entry summaries. CBP hasn’t published formal end-use certification requirements yet. Best practice is to maintain records of end-use intent — purchase orders, deployment documentation, data centre registration.
Phase 2 is a real and unresolved risk. Proclamation 11002 directed Commerce and USTR to report back within 90 days — the April 14, 2026 deadline has passed and scope remains unconfirmed. If you’re building a multi-year GPU procurement roadmap, include a Phase 2 sensitivity scenario.
Here’s what to do now: confirm your use case maps to an exempt HTSUS category; brief your finance and legal teams on the startup-exemption ambiguity; request HTSUS classification documentation from your supplier; build a Phase 2 sensitivity scenario into your hardware budget.
For a complete view of how this tariff affects your GPU procurement budget, our AI chip tariffs and semiconductor geopolitics framework covers cost scenarios end-to-end. Hyperscaler-specific exemptions are covered in the US-Taiwan $500 billion deal and who actually benefits.
Common questions about how the tariff works in practice are addressed below.
Frequently Asked Questions
Does the 25% AI chip tariff mean US companies will pay more for Nvidia H200s?
For most US companies: no. Domestic data centre, R&D, startup, and industrial uses qualify for use-based exemptions under HTSUS 9903.79.03–09. The tariff only applies when chips are imported for re-export to China. The indirect risk: if your supplier embedded an unapplied tariff cost in the unit price, you pay it. Request HTSUS classification documentation from your suppliers.
Does the tariff apply to GPUs bought from a US reseller, or only to direct imports?
The tariff applies at the point of importation — the importer of record files the HTSUS classification, not the final buyer. If a reseller imported under a use-based exemption and resells to you for the same use, no additional liability arises. If they imported without an exemption, that cost passes through to you. Supply chain transparency is your protection here.
Is the Nvidia H100 covered by the 25% tariff?
No — the H100 falls below the TPP and DRAM bandwidth thresholds, classified under exception code 9903.79.02. Separate BIS/EAR rules still apply to China-bound H100 exports, but domestic US procurement is unaffected by either regime.
Can Chinese companies still buy Nvidia GPUs legally?
Yes, with conditions. H200 and MI325X are licensable under BIS review, subject to the four criteria and the 50% volume cap. H20 exports are currently permitted without tariff. Blackwell B200 exports are categorically prohibited — no licence pathway.
What is the difference between the Section 232 tariff and a chip export ban?
An export ban prohibits a transaction entirely — no legal pathway at any price. A tariff imposes a cost penalty but allows the transaction. A ban removes market access; a tariff prices it. For procurement planning: a ban creates binary uncertainty; a tariff creates price risk you can model.
What is the “startup exemption” and does my company qualify?
Proclamation 11002 lists “startups” and “emerging growth companies” as exempt. As of April 2026, no formal definition or threshold has been published by CBP — a live regulatory gap. Maintain contemporaneous records and monitor CBP guidance.
What are Phase 2 tariffs, and should they factor into GPU procurement planning?
Phase 2 is the second Section 232 tranche, directed for assessment by April 14, 2026. It could extend tariffs to a much wider semiconductor range. The deadline has passed; scope is unconfirmed. Multi-year hardware budgets should include a Phase 2 sensitivity scenario.
Is the Section 232 chip tariff the same as the broader reciprocal tariffs?
No. Section 232 chips are explicitly excluded from IEEPA tariff stacking — no compounding duty from both regimes. Exception: the 40% IEEPA tariff on Brazil and the 10% IEEPA tariff on China do stack for chips from those origins.
Why did the US use a tariff rather than an outright export ban for the H200?
Revenue generation alongside deterrence. The tariff is also more legally durable than an emergency-powers ban and allows policy to be calibrated over time without a formal reversal. Section 232 specifically contemplates economic instruments alongside prohibitions.
How does the tariff interact with the BIS export licence process?
Separate mechanisms, both apply. BIS/EAR determines whether an export is legally permissible. Section 232 determines what duty applies if it is. Both must be satisfied for any China-bound chip moving through US territory.
What documentation does my procurement team need to claim a tariff exemption?
Importers must classify covered chips under HTSUS Chapter 99 exemption subheadings 9903.79.03–09 when filing US Customs entry declarations. CBP hasn’t published formal requirements yet. Best practice: maintain records of end-use intent — purchase orders, deployment documentation, data centre registration.
Will the 25% tariff affect the price of AI cloud services?
Indirectly and modestly. Cloud hyperscalers (AWS, Azure, GCP) are likely exempt under the data centre exemption. Smaller cloud GPU providers should verify their HTSUS classification. Any cost pass-through would be gradual — not an immediate price change.
