Commerce Secretary puts semiconductor tariff Phase Two on the record at the G20, and the exemption math threatens laptops, servers, consoles and the AI buildout itself
WASHINGTON, September 7, 2026
The Trump administration’s long-signaled second wave of semiconductor tariffs moved from anonymous deliberation to on-record policy this week, after Commerce Secretary Howard Lutnick confirmed on live television that a new round of chip duties is coming and laid out the exemption formula in a single sentence: build in the United States and pay nothing, build elsewhere and pay to enter the American market.
Speaking on CNBC’s Squawk Box and Bloomberg TV on the sidelines of the G20 Innovation Ministerial in Chapel Hill, North Carolina, Lutnick told an audience that included commerce and trade ministers from every major semiconductor-supplying nation that the administration is preparing what he described as targeted and thoughtful semiconductor tariffs. Companies that do not manufacture domestically, he said, should “expect to pay to enter the greatest market in the world.” The remarks, first delivered September 2 and reverberating through trade channels and equity markets into the weekend, represent the most direct public commitment yet from the Cabinet official with authority to implement the policy.
The confirmation ends a week of speculation triggered by an August 27 Politico report, based on eight sources familiar with internal deliberations, that the administration was weighing a dramatic expansion of its January tariffs on advanced chips. Asked directly on Bloomberg TV whether that report was accurate, Lutnick said yes. What the anonymous sourcing could not establish, and what the Commerce Secretary’s appearances did, is that the administration has crossed from weighing a policy to announcing one.
From Narrow Phase One to Sweeping Phase Two
The first phase of the semiconductor tariff program arrived on January 15, 2026, when President Trump signed Proclamation 11002 under Section 232 of the Trade Expansion Act of 1962. That action imposed a 25 percent ad valorem duty on a deliberately narrow set of advanced artificial intelligence accelerators. The accompanying White House fact sheet named Nvidia’s H200 series and AMD’s Instinct MI325X as in-scope products, and the proclamation itself labeled the action Phase 1.
Just as important as what Phase One taxed was what it spared. The proclamation carved out six exemption categories covering chips destined for US data centers, research and development, AI startups, repairs, non-data-center consumer and industrial applications, and public-sector uses. Those carve-outs were a deliberate structural decision that allowed the domestic AI infrastructure buildout to continue without tariff drag while Washington negotiated bilaterally with Taiwan, South Korea and Japan.
Phase Two, as Lutnick described it, collapses that distinction. The scope under consideration is far broader than Phase One’s, reaching beyond bare chips to finished goods that contain them. According to reporting confirmed by Tom’s Hardware and based on the same eight sources cited by Politico, potential targets include data center servers, laptops and gaming hardware, product categories that were explicitly sheltered under the January exemptions. The administration has also considered setting separate tariff rates and quotas for individual countries, with country-specific guidance covering their major semiconductor manufacturers.
Three critical details remain unspecified. Lutnick declined to name tariff rates for Phase Two, declined to say which exemption categories, if any, would survive from Phase One, and offered no timeline between announcement and implementation. The Commerce Department’s data-center semiconductor market report, mandated by Proclamation 11002 with a July 1 delivery deadline, was completed before September but has not been made public, and no explanation for the delay has been offered.
The Pharmaceutical Template
Lutnick was explicit about the model the administration intends to follow. “We gave tariff relief to companies who built their innovative pharmaceuticals in America and did MFN,” he said, referring to the Section 232 pharmaceutical tariff structure that pairs punishing headline rates with steep discounts for companies that sign onshoring agreements with the Commerce Department. “So that’s what you should look for in semiconductors.”
Under the framework Lutnick favors, tariff-free chip import allowances would be tied proportionally to a company’s investment in US semiconductor manufacturing capacity. The mechanics are already visible in the arrangement negotiated with Taiwan alongside the January proclamation. Taiwan Semiconductor Manufacturing Company, which has committed roughly 265 billion dollars to its Arizona complex in what stands as the largest foreign direct investment in US history, can import chips duty-free at tiered ratios tied to its American manufacturing capacity: 2.5 times its planned domestic output while new plants are under construction, tightening to 1.5 times output once facilities become operational.
The arithmetic of that formula is where analysts see trouble. TSMC’s own projections place only about 30 percent of its most advanced fabrication capacity in Arizona at full build-out. At the 1.5-times operational ratio, that generates duty-free coverage for roughly 45 percent of the company’s current production, leaving the majority of what American AI infrastructure operators need exposed to tariff liability even for the single most committed foreign investor in US chipmaking.
For organizations further down the food chain, the formula offers nothing at all. AI startups, university research laboratories, mid-tier cloud providers and federal research programs have no semiconductor manufacturing capacity and no capital to build any. Under a Phase Two that eliminates the January carve-outs, they would face full tariff exposure on every imported chip in a market where, by the most optimistic industry projections shared in the talks, meaningful domestic production cannot substitute for imports for at least five years.
An Awkward Split Screen in Chapel Hill
The G20 Innovation Ministerial, co-hosted by the Commerce Department and the White House Office of Science and Technology Policy, gathered ministers from all 20 member nations alongside technology executives including Nvidia chief executive Jensen Huang, OpenAI’s Sam Altman, Anthropic co-founder Tom Brown and Palantir’s Alex Karp. The summit produced a consensus statement and a set of governance commitments dubbed the Carolina Principles for Emerging Technologies.
It also produced a striking divergence. Standing in the same building where Lutnick announced the tariff direction, Huang told CNBC he had “had no conversations” at the G20 related to the forthcoming semiconductor duties, even as he praised Lutnick as being “really quite supportive of building in the United States.” Nvidia’s accelerators have been the named targets of Phase One, and its customers, including Amazon, Google, Microsoft and nearly every large AI laboratory, would bear the highest costs of Phase Two. The company at the center of the policy is conspicuously declining to co-sign it in public.
Foreign capitals responded with studied calm and quiet maneuvering. Taiwan’s president, Lai Ching-te, has characterized the island’s 20 percent reciprocal tariff rate as temporary, with semiconductors to be handled under separate sectoral rules still being negotiated. Taiwan’s rate notably exceeds the 15 percent secured by Japan, South Korea and the European Union in their trade deals. South Korea’s trade ministry, fresh from Industry Minister Kim Jung-kwan’s two-day Washington visit in August that he said produced “significant and meaningful progress,” has floated cooperation with Taipei, arguing the two chipmaking powers could jointly extract the most favorable treatment. Bilateral discussions with both governments remain active, according to industry reports.
The Legal Architecture, and Why It Is Durable
Phase Two’s legal vehicle matters as much as its rates. On February 20, 2026, the Supreme Court ruled 6 to 3 in Learning Resources v. Trump that the International Emergency Economic Powers Act does not authorize the president to impose tariffs, striking down the administration’s IEEPA-based tariff regime and forcing the termination of those duties four days later. The decision elevated Section 232, the national security tariff statute, into the administration’s primary surviving vehicle for broad import restrictions.
The February ruling was not a minor course correction. The Penn Wharton Budget Model estimated that unwinding the IEEPA duties exposed the government to as much as 175 billion dollars in potential refunds, and the administration responded by rebuilding its global program on Section 301 findings, which produced duties of 10 or 12.5 percent on imports from 60 trading partners in July, and by leaning harder on Section 232 for sectoral actions covering steel, aluminum, copper, drones, pharmaceuticals, polysilicon and now, prospectively, a much wider band of electronics. Phase Two is the largest single expansion of that post-ruling architecture yet contemplated.
The structural consequence is permanence. An IEEPA tariff could be switched off by executive order. A Section 232 tariff remains in force until a president affirmatively declares that the covered imports no longer threaten national security, and Congress cannot remove it without a veto-proof majority. Once Phase Two is finalized and published, duties on servers, laptops and gaming hardware would carry no sunset date and would survive a change of administration absent deliberate action. Trade advisers are telling clients to model the costs as a standing feature of the landscape, not a temporary disruption.
What the Economists Are Warning
The quantitative warnings arriving from research institutions are stark. The Center for Strategic and International Studies estimated in May that roughly 54 cents of every dollar spent on US data center infrastructure goes to semiconductors, and that the projected 2.7 trillion dollars in American data center capital expenditure through 2030 will require more than 1.4 trillion dollars in chips, the overwhelming majority of them imported.
The Computer and Communications Industry Association has estimated that taxing both chips and the finished products that contain them could cost the United States approximately 90 billion dollars in gross domestic product annually. The group projects that roughly 20 percent of planned data-center construction through 2030, about 450 billion dollars in capital expenditure, would be delayed, canceled or relocated abroad, threatening approximately 243,000 jobs. The Information Technology and Innovation Foundation warned in June that a blanket tariff on semiconductors and downstream goods would “ultimately weaken US growth, raise prices for consumers, and undermine American AI leadership.”
Consumers are already feeling the pre-tariff squeeze. DRAM contract prices rose 90 to 95 percent quarter on quarter in the first three months of 2026, the steepest quarterly surge in the memory industry’s recorded history. Gartner projected in February that a 130 percent surge in combined DRAM and solid-state drive prices by year-end would push personal computer prices up 17 percent and smartphone prices up 13 percent from 2025 levels, and the firm’s analysts expect the sub-500-dollar PC tier to vanish by 2028 regardless of tariff policy. The hardware market has moved in the same direction: Nintendo raised the Switch 2 to 499.99 dollars effective September 1, Microsoft’s Xbox Series X hit 799.99 dollars on August 1, and Sony’s PlayStation 5 retails at 549.99 dollars. Phase Two duties on finished electronics would compound increases already in motion.
The Equipment Chokepoint Nobody Is Discussing
A quieter vulnerability sits inside the Phase Two framework: the zero-for-zero equipment exemption that currently shields ASML, the Dutch company that is the world’s only maker of extreme ultraviolet lithography machines. Every advanced chip produced at TSMC, Samsung, Intel, SK Hynix and Micron at process nodes below roughly 7 nanometers depends on ASML’s EUV scanners. Canon and Nikon both abandoned EUV development years ago. ASML builds approximately 50 to 60 of the machines per year, a production rate that is the physical ceiling on how fast the world can add leading-edge fabrication capacity, whatever the tariff incentives.
Under the July 2025 EU-US trade deal, semiconductor manufacturing equipment was carved out of the 15 percent baseline duty on European goods. A Phase Two action under Section 232 could override or complicate that arrangement. ASML withdrew its 2026 revenue growth guidance mid-year, citing tariff uncertainty alongside reduced Chinese demand. Chief financial officer Roger Dassen has previously estimated that a 30 percent tariff on European goods would raise the price of a single high-end EUV machine from 250 million to 325 million euros, roughly 40 to 75 million dollars per scanner in added cost for the very companies the administration wants building American fabs.
The December 31 Double Bind
Hanging over the entire policy is a deadline Congress has not addressed. The Advanced Manufacturing Investment Credit under Section 48D, a 35 percent refundable tax credit for qualified investment in US semiconductor facilities, requires construction to begin by December 31, 2026. Any fab that has not broken ground by that date loses eligibility entirely. The industry association SEMI brought senior tax executives to Capitol Hill on July 22 to press for a multi-year extension, and Senate Finance Committee Chairman Mike Crapo of Idaho and Ranking Member Ron Wyden of Oregon issued a bipartisan statement of support on August 5. No legislation has passed.
The bind is precise. Phase Two raises the cost of importing at the same moment the incentive that offsets the cost of building domestically may lapse. Constructing a leading-edge fab in the United States currently runs 30 to 50 percent more than in Taiwan, South Korea or Singapore, a gap CSIS attributes primarily to rival governments’ incentive regimes. South Korea offers research and development incentives of 30 to 50 percent, Taiwan offers credits of up to 35 percent for advanced nodes plus duty exemptions on specialized equipment, and Japan provides a 20 percent corporate income tax reduction alongside targeted grants. Without the American credit, the cost gap widens exactly when the tariff makes domestic investment most urgent.
Can Tariffs Outrun Physics?
The administration’s stated objective is to onshore chipmaking by making imports expensive enough that foreign producers build American fabs to escape the duties. Industry representatives have raised a specific and quantifiable objection in private talks: even the most aggressive US investment program cannot close the near-term gap Phase Two would create.
The timelines tell the story. SK Hynix broke ground on its 4-billion-dollar-plus advanced packaging facility at Purdue Research Park in West Lafayette, Indiana on August 27, in a ceremony attended by Governor Mike Braun and Senator Todd Young. The company’s own schedule puts the cleanroom opening in October 2028 and mass production of next-generation HBM4E memory in the second half of 2029, meaning the American operation will depend on Korean wafer production for years regardless of tariff levels. US fabs take more than three years on average from permitting to first production, roughly a year and a half longer than equivalent Taiwanese plants. The high-bandwidth memory market that feeds AI accelerators is even tighter: SK Hynix controls about half of it, Samsung roughly 30 percent, and US-based Micron most of the remainder, with supply sold out through 2026 before Phase Two was even confirmed.
Winners, Losers and the China Question
Running alongside the chip escalation is a parallel action aimed at the materials layer of the same supply chain. On August 6, President Trump signed a proclamation imposing a 15 percent ad valorem tariff on downstream polysilicon derivatives, including ingots, wafers, solar cells and solar modules, together with minimum import prices on all polysilicon-derived products, effective December 4. The price floors were set at 21 dollars per kilogram for polysilicon, 100 dollars per kilogram for ingots and wafers, 22 cents per watt for solar cells and 38 cents per watt for solar modules, according to securities disclosures filed by affected utilities. Because China dominates global polysilicon production, the measure functions as a direct pressure point on Chinese supply chains for both semiconductors and solar panels, extending the industrial policy logic of the chip program into raw materials.
The distributional consequences of Phase Two, if finalized as described, would be uneven in ways the administration has not publicly addressed. Intel and Micron, with substantial existing American fabrication capacity, would enjoy structural advantage over import-dependent rivals, and both companies’ shares have outperformed the sector since the Politico report surfaced. Contract manufacturers and original design manufacturers in Taiwan, Vietnam and Mexico that assemble laptops and servers for the US market face classification questions worth billions: whether a finished notebook is tariffed on its full value or on its semiconductor content is a design detail that will determine the economics of entire product categories. Retailers with long inventory cycles, including the big-box chains that place holiday electronics orders in late summer, made their 2026 season commitments before the September 2 confirmation and now face repricing decisions mid-cycle.
There is also a demand-side wildcard: hyperscaler capital budgets. Amazon, Microsoft, Google and Meta have collectively guided to hundreds of billions of dollars in 2026 and 2027 data center spending. Those budgets were built on chip cost assumptions that predate Phase Two. If the duties push accelerator costs up materially, the companies face a choice between absorbing the increase, which compresses margins, or slowing the buildout, which is precisely the outcome the CCIA jobs and GDP warnings describe. Analysts at several banks began publishing Phase Two sensitivity scenarios for the hyperscalers within a day of Lutnick’s remarks.
For importers, purchasing managers and compliance teams, the practical guidance emerging from trade counsel this week is to treat Phase Two as a near-certainty with unknown parameters. Companies with exposure to servers, laptops, consumer electronics or AI infrastructure should be modeling scenarios at multiple rate levels, auditing country of origin and Harmonized Tariff Schedule classifications for semiconductor content, and evaluating whether accelerating shipments ahead of a proclamation is feasible. Hyperscalers and startups alike now have an on-record commitment from the official with the authority to implement the policy, and the message from Chapel Hill was unambiguous: the administration is prepared to absorb a near-term supply squeeze in exchange for long-term manufacturing leverage, and the bill will land on every company, and every household, that buys the products chips go into.