US Commerce Secretary Howard Lutnick is advancing a new tariff framework aimed at pulling more semiconductor manufacturing onto US soil, with early reporting suggesting the scope could widen from advanced AI chips to AI servers, laptops and gaming consoles.
Commerce Secretary Howard Lutnick is seeking a new tariff round to accelerate semiconductor manufacturing's return to the United StatesCITE:E1. The move builds on an existing measure: in January 2026, the Trump administration used Section 232 authority to impose a 25% tariff on select advanced computing chips, including NVIDIA's H200 and AMD's MI325X, while carving out exemptions for US data centers, R&D, startups and repair usesCITE:E3.
The US Commerce Department is weighing a plan that ties a company's tariff-free chip import quota to how much manufacturing capacity it commits to build inside the United States, with larger domestic production pledges unlocking larger duty-free allowancesCITE:E2. POLITICO, citing unnamed sources, reported on August 27 that Lutnick favors linking tariff relief to foreign companies' investment in US chip manufacturing and is considering a grace period before any new tariff mechanism takes effectCITE:E6. Tariff rates and other details remain undecided, and the policy could still change substantially over the coming weeks or monthsCITE:E6.
Taiwan produces more than 90% of the world's most advanced semiconductors, leaving the United States dependent in the near term on supply chains centered in Taiwan, South Korea and MalaysiaCITE:E7.
TSMC has committed $265 billion to its Arizona operations covering advanced process nodes, advanced packaging and R&D, but even once every planned phase is complete, the company estimates only about 30% of its most advanced capacity will be located in the United StatesCITE:E4CITE:E8.
Building a fab from groundbreaking to volume production takes several years, leaving US advanced-node supply unable to fully absorb fast-growing AI demand in the near termCITE:E5. A trade and technology figure who regularly meets senior Trump administration officials estimates that establishing domestic US chip manufacturing capacity will take more than five yearsCITE:E9.
A technology industry representative says the tariff-free import quotas under discussion would fall short even for hyperscale cloud providers alone, let alone the rest of the industryCITE:E10. "The math just doesn't work. What they're talking about for duty-free import quotas isn't even enough for the hyperscalers, let alone other industries. We can't source these chips in the US because the capacity we'd need doesn't exist yet," the representative saidCITE:E10.
| Metric | Value | Source |
|---|---|---|
| Section 232 tariff on select advanced computing chips (Jan 2026) | 25% | CITE:E3 |
| TSMC's committed Arizona investment | $265 billion | CITE:E4 |
| Taiwan's share of world's most advanced chip production | >90% | CITE:E7 |
| TSMC's most advanced capacity expected in the US after full build-out | ~30% | CITE:E8 |
| Estimated time to build domestic US chip manufacturing capacity | 5+ years | CITE:E9 |
What this means: Lutnick's proposed design ties tariff relief to US manufacturing commitmentsCITE:E2CITE:E6, yet the two companies most central to that buildout math show the size of the gap — TSMC's own $265 billion Arizona commitment still leaves only about 30% of its most advanced capacity in the US even at full completionCITE:E4CITE:E8, fab construction runs several years to volume production and an industry source puts full domestic capacity buildout at more than five yearsCITE:E5CITE:E9, and a technology representative says the proposed duty-free quotas already fall short of hyperscaler needs aloneCITE:E10 — all while Taiwan continues to supply more than 90% of the world's most advanced chipsCITE:E7.
Commerce Secretary Howard Lutnick is seeking a new tariff round to accelerate semiconductor manufacturing's return to the United StatesCITE:E1.
The US Commerce Department is weighing a plan that ties a company's tariff-free chip import quota to how much manufacturing capacity it commits to build inside …
Taiwan produces more than 90% of the world's most advanced semiconductors, leaving the United States dependent in the near term on supply chains centered in Tai…
TSMC has committed $265 billion to its Arizona operations covering advanced process nodes, advanced packaging and R&D, but even once every planned phase is comp…
The core tension in this policy design is timing. Lutnick's plan to size tariff-free quotas to US manufacturing commitments assumes domestic capacity can flex with policy, but TSMC's own $265 billion Arizona buildout — even fully realized — still leaves only about 30% of its most advanced capacity in the US, against fab timelines industry sources put at five-plus years to volume production. That mismatch is precisely why a technology representative already says the proposed duty-free allowance wouldn't cover hyperscalers alone. The detail worth watching is whether the 'grace period' POLITICO's sources described survives into a final rule: a multi-year buffer would at least acknowledge the physical build-out math; a short one would collide directly with the fact that Taiwan still supplies more than 90% of the world's most advanced chips today.
SK Hynix (SK海力士) is planning a ₩10 trillion plant investment in Japan, a move that coincided with gains in two Japan-focused semiconductor ETFs. The plan follows a first half of 2025 in which the company posted ₩39.87 trillion in revenue, ₩15.10 trillion in net profit, and a debt-to-equity ratio that fell to 48.13% from 62.15% at the end of 2024.
Taiwan's benchmark TAIEX closed at 46,331.45 points on August 28, up 356.23 points and above the 46,000 threshold, on turnover of roughly NT$1.023 trillion. TSMC rose NT$10 to NT$2,420, and 36 listed stocks hit their daily limit after NVIDIA's earnings beat lifted sentiment at the open.
Four hyperscalers are on track to spend roughly $725 billion combined on AI infrastructure in 2026, up about 77% from 2025's $410 billion, and NVIDIA's data center revenue confirms real near-term demand with 117% year-over-year growth. Yet Bain projects an $800 billion annual revenue shortfall by 2030, Michael Burry alleges $176 billion in understated depreciation, and circular deals linking NVIDIA, OpenAI, and Oracle draw dot-com comparisons — even as NVIDIA and JPMorgan Asset Management reject the bubble framing.