Phase 2 Chip Tariffs Would Hit Laptops, Consoles — and AI Data Center Servers

Just as the AI industry was digesting January’s 25% tariff on advanced accelerators, the Trump administration is preparing something broader. According to a POLITICO report published August 27 and confirmed by eight unnamed sources familiar with the talks, Commerce is weighing a second round of semiconductor tariffs — “Phase 2” — that would extend duties beyond raw chips to the finished products built with them: laptops, gaming consoles, and, most consequentially for the AI economy, data center servers.

What Phase 2 Would Do

The reporting, corroborated by CNBC and picked up across the hardware press, describes a framework championed by Commerce Secretary Howard Lutnick with a simple but radical mechanic: duty-free import quotas pegged to each company’s committed US manufacturing capacity. Companies could import a certain volume of chips — and possibly finished systems — tariff-free, with the allowance scaled to how much production they pledge to bring to the United States. Import more than your commitment justifies, pay the tariff.

Key details remain unsettled. The administration is reportedly considering a phase-in period, while tariff rates and terms have yet to be determined. One proposal would establish separate tariff rates and import quotas for individual countries, along with country-specific guidance covering the major semiconductor manufacturing nations.

The most alarming signal for the tech industry: Commerce officials have indicated in private talks that the exemptions attached to January’s tariff may not carry over. Proclamation 11002, signed January 14, imposed a 25% duty on a narrow set of advanced accelerators — Nvidia’s H200 and AMD’s MI325X were named in the accompanying White House fact sheet — but carved out broad exemption categories covering US data centers, research and development, startups, repairs, non-data-center consumer and industrial applications, and public sector uses. Those carve-outs are exactly what’s now in question.

The Math Doesn’t Work Yet

The core tension is arithmetic. Taiwan produces more than 90% of the world’s leading-edge chips. Under Taiwan’s January trade agreement, Taiwanese chips enter at zero tariffs within 2.5 times a company’s current US manufacturing capacity while new plants are under construction, tightening to 1.5 times once they’re built. TSMC has committed $265 billion to its Arizona site — the largest foreign direct investment in US history — yet even at full build-out projects only around 30% of its most advanced capacity there.

Industry representatives arguing in the talks made the obvious point: a quota keyed to current domestic capacity cannot cover the volumes hyperscalers are buying during a record AI spending run. Jonathan McHale, digital policy chief at the Computer and Communications Industry Association, whose members include Amazon, Google, and Meta, compared the data center buildout to “building the transcontinental railroad” and warned that added cost and unpredictability put that investment at risk.

One person involved in the discussions put the realistic timeline for the domestic manufacturing build-out at more than five years — longer than any phase-in period the administration has allowed on previous tariff rounds. Tech lobbyists have met with Lutnick and Bureau of Industry and Security undersecretary Jeffrey Kessler with growing frequency since the start of summer, but three sources said recent talks have moved against the industry. The White House, for its part, is unmoved: spokesperson Kush Desai defended the approach, saying reshoring chip manufacturing is a top priority for the president.

Why Servers Are the Stakes

For the AI industry, the inclusion of data center servers in the potential tariff scope turns a consumer electronics story into an infrastructure problem. TrendForce’s analysis of the POLITICO report notes that chips manufactured by TSMC in the US are already tariff-exempt — but most chips are shipped to assembly plants in China and Southeast Asia, incorporated into laptops, servers, and consoles, and only then exported to the US. Tariffs on those finished products would weaken demand and ripple back upstream to semiconductor manufacturers.

The impact would be uneven. AI servers may see relatively limited effects, TrendForce suggests, because their high value has already prompted manufacturers to expand US production. PCs face greater pressure: average prices are already up more than 20% in the first half of 2026 due to surging memory and component costs, and additional tariffs on laptops would almost certainly be passed through to consumers, further weighing on demand.

The Memory Squeeze Compounds Everything

The most perverse wrinkle: the tariff threat lands on top of the worst memory supply crunch in years. Global supplies of DRAM, NAND flash, and HBM remain severely constrained, with significant new supply not expected until 2028 despite capacity expansions by Samsung, SK hynix, and Micron. With memory prices already surging, tariffs could shift the pressure onto US Big Tech companies that need enormous memory volumes to build AI data centers — firms that, in many cases, have no alternative supplier to switch to.

The geopolitics of memory investment cuts both ways. SK hynix held a groundbreaking ceremony for its West Lafayette, Indiana packaging plant on August 27 — but unlike Samsung, it has yet to make a major US investment in front-end chip production, raising the possibility of further announcements. Yet expanding now exposes the company to oversupply risk if the semiconductor cycle turns down. Tariffs designed to force reshoring are colliding with an industry that remembers what overcapacity looks like.

What Happens Next

Nothing is final. The framework “could still change substantially in the coming weeks,” per the report, and the phase-in discussion suggests the administration is aware of the disruption a hard cutoff would cause. But the direction of travel is clear: January’s exemptions were the sugar that helped Phase 1 go down, and Commerce is now signaling that Phase 2 may not be so sweet.

For AI infrastructure buyers, the practical takeaways are straightforward. First, the era of assuming hardware cost curves only improve is over — tariff risk is now a line item in data center planning. Second, companies with committed US manufacturing footprints (TSMC’s Arizona partners, onshore server assemblers) may gain a meaningful cost advantage over import-reliant rivals. Third, the five-year build-out timeline means any gap between quota and demand lands on exactly the companies racing to stand up AI capacity now.

The transcontinental railroad analogy is more apt than its author may have intended. That project succeeded because the government made it cheap to build. Phase 2, as described, would make it expensive to build anything else.

---

Sources are listed in the article metadata. Key reporting: POLITICO (Aug 27, 2026, eight sources), TrendForce analysis citing UDN News, Chosun Ilbo, and Hankyung (Aug 28, 2026), and Tom’s Hardware policy coverage (Luke James, Aug 27, 2026).