White House Plans New Comprehensive Tariffs on Chips
The U.S. government is evaluating expanding semiconductor tariffs from the current narrow scope to a more comprehensive coverage, potentially including chips, some manufacturing equipment, and downstream chip-containing products. The Department of Commerce has completed the Section 232 investigation, and the White House previously clarified that the first phase imposes a 25% tariff only on specific advanced computing chips, with a "significant rate" possible after the conclusion of the second phase negotiations.
The existing 25% tariff targets advanced logic chips and chip-containing products that meet specific computing power and bandwidth parameters, typically aimed at products like NVIDIA's H200 and AMD's MI325X, and does not apply to imports used in U.S. data centers, R&D, startups, civilian industries, consumer electronics, and the public sector. The White House also retains the power to impose tariffs on a broader range of chip categories and plans to implement "tariff offsets," tying the reduction amount to domestic capacity investments in the U.S.
The President previously stated that tariffs on chip imports that have not committed to building factories in the U.S. could rise from about 25% to higher rates, even mentioning around 100%. Commerce Secretary Howard Lutnick gave memory manufacturers a choice at Micron's New York factory groundbreaking event: pay a 100% tariff or relocate production to the U.S. Samsung and SK Hynix are the main players in global HBM, while Micron is expanding high-bandwidth memory production in Idaho and New York.
Tech and business organizations have pressured the Department of Commerce and the Office of the U.S. Trade Representative. The Consumer Technology Association, National Foreign Trade Council, and the Electrical Manufacturers Association pointed out that tariffs on chips and downstream complete machines would raise costs for data centers and end products, slowing down the expansion of artificial intelligence in the U.S. The Center for Strategic and International Studies estimates that without targeted exemptions from the 232 tariffs, U.S. data center construction would be significantly impacted, as the U.S. currently consumes about a quarter of the world's chips but only produces about one-tenth.
The supply chain remains concentrated in East Asia. TSMC handles the majority of advanced processes globally and has committed to additional large-scale investments in the U.S.; discussions between the U.S. and Taiwan have considered about $250 billion in related investments, with import exemptions based on domestic capacity. The combination of memory shortages and artificial intelligence training demand has already driven up prices for HBM and DRAM, and comprehensive tariffs would directly increase costs for servers, smartphones, and automotive electronics.
From a market mechanism perspective, this is a policy-driven forced relocation of factories, not a sudden disappearance of demand. Buyers are cloud providers, server assembly plants, and consumer electronics brands that must import advanced processes and memory; sellers are foreign foundries and packaging/testing capacities like TSMC, Samsung, and SK Hynix. The flow of funds is tariff revenue into federal finances while forcing capital expenditures to lock in U.S. foundries, packaging plants, and equipment orders. Beneficiaries include Intel, Micron, and cloud providers that have already invested in the U.S. and received exemption quotas; those under pressure are complete machine manufacturers still relying on overseas advanced processes and AI model companies passing on tariff costs to training expenses.
Source: Public Information
ABAB AI Insight
During Trump's first term, Section 232 was used to leverage negotiations over steel and aluminum, and in the second term, the same statute was applied to chips. When the semiconductor Section 232 investigation was initiated in April 2025, the official narrative was "reviving key technology manufacturing"; however, the implementation in January 2026 resulted in a narrow 25% tariff specifically targeting high-end AI chips that could be re-exported to China, effectively imposing a transit tax first and reserving comprehensive tariffs as a second-phase threat. NVIDIA relies on H200 sales to China for licensing, and Jensen Huang has become one of the closest tech lobbying channels to the White House, indicating that tariffs have shifted from industry protection to tradable licenses.
The capital pathway is "tariffs in exchange for building factories." TSMC has committed to advanced capacity as part of U.S. investment, in exchange for tax-free imports at 2.5 times the planned capacity during the construction period; on the memory side, Lutnick has directly targeted Samsung and SK Hynix with a 100% tax rate, pushing for high-bandwidth memory to be established in the U.S. The CHIPS Act is about providing subsidies to attract people, while this round is about raising tariffs to compel them to come. The motivation is not to immediately replace Taiwan's 12-inch advanced processes but to lock wafer investments, equipment orders, and packaging/testing into U.S. balance sheets while using tariff revenue to compensate federal finances.
Comparative examples include the trade war against Japanese semiconductors in the 1980s and the Section 232 tariffs on steel and aluminum after 2018. The U.S. pressured Japan's capacity share through anti-dumping measures and the Plaza Accord, resulting in Korea and Taiwan taking over, with U.S. domestic manufacturing share dropping from about 40% to 10%. If comprehensive tariffs are imposed on advanced processes today, it may temporarily raise prices like steel and aluminum, but in the long term, it could shift packaging/testing and mature processes back, while processes below 3nm will still rely on TSMC. The industry is in a phase of "coerced expansion": technological substitution is not complete, but market access is being exchanged for geographical relocation.
Structural judgment belongs to the reconstruction of the industrial chain. The pricing power of artificial intelligence computing power does not lie with terminal brands but with advanced processes and HBM supply. Comprehensive tariffs increase the dollar cost of imported chips, effectively raising the marginal cost of training clusters; whoever secures U.S. exemption quotas first will maintain the speed of computing power expansion. The mechanism is: what the U.S. lacks is manufacturing, not design; tariffs change the crossing price of chips but do not alter extreme ultraviolet lithography and yield curves. The result is that cloud providers accelerate stockpiling, foreign foundries expedite U.S. construction, while China continues to fill gaps with mature processes and domestic alternatives.
ABAB News · Cognitive Laws
Tariffs are not walls; they are invoices compelling the other party to move factories.
In computing power competition, the loss is in manufacturing, not in slogans.
Exemptions are worth more than tax rates; quotas are the real tariffs.