NVIDIA designs all its GPUs in the US but manufactures them exclusively at TSMC in Taiwan. This makes Taiwan the country of origin for US customs. NVIDIA H100/H200 AI accelerators face the 10% Section 122 surcharge on their $25,000–$40,000 price points, resulting in significant per-unit duty costs.
NVIDIA is headquartered in United States, but most components are manufactured in Taiwan, China. For US customs purposes, the COO is where the component is manufactured — not where the company is based. Declaring United States as COO when parts ship from Taiwan is a customs violation.
| Country | Sec 122 | Sec 301 | Total |
|---|---|---|---|
| Taiwan | 10% | — | 10% |
| China | 10% | 50% | 60% |
* Section 301 rates shown for electronics HTS chapters (8541–8542). Rates vary by product.
| HTS Code | Description |
|---|---|
| 8542.31.0000 | GPU chips (H100, H200, A100) |
| 8473.30.1180 | GPU boards and cards |
NVIDIA operates as a pure-play fabless company: all chip design occurs at headquarters in Santa Clara, California, but all wafer fabrication is outsourced to TSMC in Taiwan. The H100, H200, A100, and B200 AI accelerator chips are manufactured at TSMC's Hsinchu and Tainan fabs using 4 nm (N4) and 3 nm (N3E) process nodes. The completed GPU dies are then packaged by ASE Group or SPIL in Taiwan using CoWoS (Chip on Wafer on Substrate) advanced packaging for HBM integration. Final GPU board assembly (PCIe add-in cards, SXM modules) occurs at various ODM facilities in Taiwan and China. For raw GPU die imports and SXM modules, Taiwan is the primary declared COO. Board-level assemblies may carry Taiwan or China COO depending on final assembly location.
NVIDIA GPU chips (bare die or packaged ICs) classify under HTS 8542.31.0000 (electronic integrated circuits — processors). GPU boards and PCIe accelerator cards classify under HTS 8473.30.1180 (parts and accessories for ADP machines) or potentially 8542.31.0000 depending on how they are invoiced. Taiwan-origin NVIDIA components: 10% Section 122 surcharge, no Section 301. An H100 SXM5 module priced at approximately $32,000 list carries a Section 122 duty of $3,200 per unit — a substantial cost that cloud providers and enterprise buyers must account for in capex planning. At the scale of a 10,000-GPU cluster, Section 122 duties alone could represent $32 million in added cost beyond the GPU price.
The concentration of NVIDIA GPU manufacturing at TSMC in Taiwan creates a single-point geographic risk that is widely acknowledged in the US defense and technology policy communities. All H100, H200, A100, L40S, and B100/B200 chips depend on TSMC's CoWoS packaging capacity, which is fully committed through 2026 based on industry reports. There is currently no alternative manufacturing pathway for advanced NVIDIA GPUs. The Taiwan Section 122 tariff rate of 10% is materially lower than the China Section 301 + Section 122 rate of 60%, meaning any hypothetical shift of production to China would sharply increase the landed cost of GPUs in the US market.
NVIDIA's AI chips are subject to US Bureau of Industry and Security (BIS) export controls under the EAR, with the H100, H200, and A800 included in controlled categories requiring export licenses for shipment to certain countries including China. For US importers, the primary compliance obligation is accurate HTS classification and COO declaration on CBP Form 7501. However, re-export compliance is a parallel requirement: entities purchasing NVIDIA AI chips for subsequent sale or transfer must review EAR §742.6 and the Entity List prior to any transaction. The combination of import duty (10% on Taiwan COO) and export control licensing requirements makes NVIDIA AI accelerator procurement one of the more compliance-intensive categories in electronics sourcing.
NVIDIA's Blackwell architecture (B100, B200, GB200) is manufactured using TSMC's 4NP process in Taiwan and assembled using NVLink Switch systems and NVL72 rack-scale configurations. The per-rack cost of a GB200 NVL72 system exceeds $3 million at list price. At 10% Section 122, the duty on a single rack import would exceed $300,000. Data center operators and system integrators should model Section 122 duty into their total cost of ownership calculations and consult with licensed customs brokers regarding the classification of complete rack-scale AI systems (which may qualify for different HTS treatment as complete systems versus component-level entries).
Enter any NVIDIA part number to get the COO, HTS code, and full duty calculation including MPF and HMF.
Look Up NVIDIA Part →