Every hardware program kickoff covers the same ground. Schedule, budget, headcount, spec. Someone asks about component availability. Someone asks about the test plan.
Almost nobody asks what the product will owe at the border.
That question used to belong to finance. As of 2026, it belongs to hardware engineering, because the decisions that set duty exposure happen at the design review, not at the shipping dock.
The rules changed in January, and they are still changing
On January 14, 2026, the White House issued Proclamation 11002, imposing a 25 percent Section 232 tariff on a defined set of advanced semiconductors, effective January 15. The White House fact sheet named NVIDIA H200 and AMD MI325X devices as examples of covered products, and carve-outs for data centers, R&D, and startups kept the practical impact small. Most hardware teams shrugged.
That was a mistake, because the January action was explicitly labeled Phase 1.
The proclamation built in two checkpoints: a negotiation update due April 14, and a report on the semiconductor market for US data centers due to the President by July 1, the trigger for deciding whether to modify the tariffs. That deadline has passed, and the decision now sits with the President. The expected next step is a Federal Register notice proposing the Phase 2 scope.
What Phase 2 would cover
The Commerce Secretary recommended that the Phase 2 tariff be set at a significant rate of duty, accompanied by a tariff offset program giving preferential treatment to companies investing in US semiconductor production.
Analysts expect the expansion to reach semiconductor manufacturing equipment, memory chips including DRAM and NAND, analog and mixed-signal integrated circuits, and derivative products: finished goods containing covered semiconductors, potentially including servers, networking equipment, and AI inference hardware.
Read that last category again. Derivative products means the finished device. If your product ships with a covered chip inside, the duty may apply to your product, at full customs value. One more detail that surprises teams: the proclamation blocks duty drawback, so re-exporting later does not get the money back.
The market is already pricing this in. NVIDIA has raised prices on AI GPUs by 5 to 15 percent, and Deloitte’s 2026 semiconductor outlook projects memory prices rising 30 to 50 percent through mid-2026.
Your BOM is a customs document
Two products with identical function can land with very different duty bills. A stamped aluminum enclosure and a molded polymer one sit in different HTS (Harmonized Tariff Schedule) territory, and with the 50 percent metals tariffs in play, the gap between them can run tens of duty points depending on classification. A device built on an advanced logic SoC (system on chip) covered by Section 232 carries exposure that the same device on a mature-node part avoids entirely.
This is why HTS classification belongs in the DFM (design for manufacturing) review, next to tolerance analysis. Metal-content duty is a legitimate input to the enclosure material trade, the same way MOQ (minimum order quantity) and tooling cost are. The analysis takes a few days. It routinely moves the landed-cost answer more than the factory quote does.
Country of origin is engineered, not declared
Vietnam exported 165 billion dollars in electronics last year. Underneath that number, roughly 80 percent of components are imported and over 90 percent of tier-1 suppliers are foreign-owned.
US Customs applies a substantial transformation test, and final assembly of near-complete Chinese kits often fails it. A “Made in Vietnam” label you cannot defend is a duty bill waiting to arrive.
Vietnam exported 165 billion dollars in electronics last year. Underneath that number, roughly 80 percent of components are imported and over 90 percent of tier-1 suppliers are foreign-owned.
US Customs applies a substantial transformation test, and final assembly of near-complete Chinese kits often fails it. A “Made in Vietnam” label you cannot defend is a duty bill waiting to arrive.
Nearshoring to Mexico gets pitched as the clean alternative, and it has real advantages: USMCA duty-free treatment for qualifying goods, ocean freight replaced by truck, transit measured in days instead of weeks, and overlapping time zones for engineering support. But the same origin logic applies. USMCA preference depends on meeting rules of origin, with regional value content thresholds that a product full of Asian components can fail. A Mexican plant doing final assembly on imported kits faces the same substantial transformation question as a Vietnamese one. And with the USMCA review underway this year, the rules themselves are a moving target worth watching before committing tooling.
The fix is the same everywhere: a subassembly origin map for every candidate factory, whether it sits in Hai Phong or Tijuana. Where each major piece is actually made, placed next to the HTS classification and the applicable rules of origin. Days of work. It settles arguments that otherwise run for months.
Why timing beats forecasting
Here is the part that matters for anyone with a 2027 launch on the roadmap.
A product entering development today reaches production in 12 to 24 months. Component selection, classification, and manufacturing location get locked at the architecture stage. Changing them at DVT (Design Validation Test) costs multiples of what the same decision costs at concept.
Nobody can forecast exactly where Phase 2 lands. You do not need a forecast. You need architecture-stage decisions that survive multiple outcomes:
- Check processor candidates against current Section 232 definitions before the architecture freezes.
- Run HTS classification on every candidate design, including the enclosure.
- Qualify a second source at PVT (Production Validation Test), while the program still has engineering attention. Malaysia for semiconductor-adjacent products, Vietnam or Thailand for high-volume assembly, Mexico when USMCA qualification and transit time justify the supply base tradeoffs.
- Re-check classifications quarterly at MP (mass production). The covered definitions have already shifted, and a Federal Register notice could shift them again.
The offset program adds one more lever. Companies investing in US capacity would get preferential tariff treatment, in a structure that parallels the CHIPS Act investment tax credit. For products with US-bound volume, a low-volume US final-assembly path is a direct answer to the broadening scenario.
The bottom line
Tariff exposure in 2026 is an output of engineering decisions. Getting ahead of it costs a few days per program. Ignoring it means finding out from a customs invoice, when every option left is expensive.
If a 2027 program is on your roadmap, the cheapest version of it starts engineering this quarter.
AJProTech takes hardware from concept through mass production, with NPI in Taiwan and manufacturing partners across Asia. If you want a partner who thinks about duty exposure at the design review instead of at the customs invoice, let’s talk.
