Chip Tariffs Copy a Taiwan Deal That Falls Short

Commerce Secretary Howard Lutnick wants chip tariff relief tied to U.S. factory pledges, a formula already used on Taiwan that still leaves most advanced chips overseas. Officials are now weighing a second round of semiconductor tariffs that could reach servers, laptops, and gaming consoles.

Those duties may also drop the data-center carve-outs that have shielded the AI buildout since January.

Phase 2 Would Put Duties on Data Center Servers

People familiar with the talks say one option would widen the tax from imported chips to many goods made with them, including the servers that fill AI halls. Tech lobbyists have been in the building all summer, meeting Lutnick, Commerce Undersecretary Jeffrey Kessler, who runs the Bureau of Industry and Security, and other senior officials.

They are asking the White House to keep something close to the January design, which put a 25% tariff on certain advanced computing chips such as Nvidia’s H200 and AMD’s MI325X, then excused the uses that actually drive U.S. demand. Recent sessions have gone the other way. Three people in those rooms said officials now talk about applying the next round more broadly, to push foundries to add U.S. plants.

The January carve-outs are the live fight. They cover almost every buyer the AI boom actually has.

JANUARY USES THAT MAY NOT SURVIVE

• U.S. data centers: Chips for halls that need more than 100 megawatts of new load for AI training, inference, simulation, or synthetic data.
• Repairs and replacements: Parts brought in to fix or swap gear already on U.S. soil.
• Research and startups: Lab work and young firms that the January order also left outside the 25 percent duty.
• Consumer, factory, and public uses: Non-data-center gadgets, civil industrial gear, government systems, and other cases the commerce secretary decides help the U.S. supply chain.

Kessler has sounded more open to those warnings than his boss, people in the meetings said, but they do not expect him to overrule Lutnick. The rate itself is still unset. So is the length of any phase-in. Country-by-country quotas are also on the table. The White House says reshoring chipmaking is a top goal and that its policies have already locked in hundreds of billions of dollars of plant pledges.

Lutnick Already Road-Tested the Quota in Taiwan

The relief model Lutnick wants is not a sketch on a whiteboard. On January 15, 2026, the day after the 25 percent duty took effect, Washington and Taipei announced a trade package that does the same thing. Taiwanese firms pledged at least $250 billion of U.S. investment in chips, energy, and AI. Taipei added at least $250 billion in credit guarantees. Lutnick called the pair a $500 billion down payment.

Goods from Taiwan moved to a 15 percent rate from 20 percent. Chipmakers that expand in the United States won something more specific. Commerce said they can import up to 2.5 times the capacity they are building, with no extra semiconductor duties, during an approved construction period. After the plants open, the duty-free allowance falls to 1.5 times U.S. output. Lutnick put it in wafers: build capacity for 1 million wafers, and you can bring in 2.5 million duty-free while the site is going up.

The U.S. trade office also confirmed preferential treatment in the Section 232 chip case, citing Taiwan’s security alignment. Two-way goods and services trade with Taiwan was more than $185 billion in 2024. Lutnick has said the aim is to move 40 percent of Taiwan’s chip supply chain to America during President Donald Trump’s term. Firms that refuse the plants, he said, should expect a much harder levy.

That’s what they get if they don’t build in America, the tariff’s likely to be 100 percent.

Howard Lutnick, Commerce secretary, on CNBC

He used the same pitch after a Micron groundbreaking, telling memory makers they can pay 100 percent or build in America. That is industrial policy, he said. The Department of Commerce later restated the model in its own words: start with chips, then use the duty to pull manufacturing back.

Secretary Lutnick on the historic semiconductor agreement:

“Yesterday, we announced the Taiwan deal, which allows Taiwan to invest $500 billion in America to bring semiconductor manufacturing here. One of the key drivers of this investment is that Micron currently produces its… pic.twitter.com/vTbThSQTcn

— U.S. Department of Commerce (@CommerceGov) January 16, 2026

THREE VERSIONS OF THE SAME LEVER

Copying that quota onto the whole market is the ironic turn. The Taiwan deal was supposed to prove that investment can buy relief. It also proved that relief is rationed by plants that do not exist yet.

Arizona’s $265 Billion Still Leaves Most Advanced Wafers Overseas

Taiwan Semiconductor Manufacturing Co. is the test case, because it makes the leading-edge logic that Nvidia, AMD, Apple, and the cloud firms design but do not etch themselves. On July 16, 2026, TSMC added $100 billion for more Arizona plants, taking its U.S. pledge to $265 billion. The Arizona Commerce Authority lists 10 fabs, two advanced packaging sites, and a research center.

Even at that scale, the state’s own readout says that once every announced fab is finished, only around 30% of TSMC’s most advanced capacity will sit in Arizona. The other 70 percent stays abroad. Taiwan still produces more than 90 percent of the world’s most cutting-edge semiconductors. A March analysis from the Center for Strategic and International Studies found U.S. firms will remain tied to Taiwan through at least 2028, and that Arizona wafers still travel back across the Pacific for advanced packaging.

Chairman C.C. Wei would not lock a calendar for the newest four plants. The schedule, he told analysts, depends on the market. TSMC’s first Arizona fab has been in volume production on N4 technology since late 2024, with yields the company compares to Taiwan. The second fab, on 3-nanometer, is built and is due to start volume in 2027. A third, aimed at 2-nanometer and A16, is still a late-decade job. The campus employs more than 3,500 people. Wei said in June it will be a long time before TSMC can meet customer demand. Chief financial officer Wendell Huang has flagged a shortage of construction workers and of local infrastructure.

THE ROAD FROM THE 232 REPORT TO ARIZONA’S CEILING

1. December 22, 2025: Commerce sends the president its Section 232 report on chip, tool, and derivative imports.
2. January 14, 2026: The president signs the semiconductor proclamation, sets a 25 percent duty on a narrow set of advanced chips, and describes a later, broader tariff plus an offset for firms that build in the United States.
3. January 15, 2026: The 25 percent duty takes effect at 12:01 a.m. eastern time, and the Taiwan package sets the 2.5 times construction quota.
4. July 16, 2026: TSMC lifts its Arizona bet to $265 billion and still projects only about 30 percent of 2-nanometer and more advanced output there at full build-out.
5. Late August 2026: Officials weigh dropping the January carve-outs and extending duties to servers and other finished machines.

One person who meets senior officials on this file put the U.S. advanced-chip build at more than five years. That is longer than the phase-in windows Trump has used on other tariffs. “The math literally just does not work,” the tech representative said. “The volume they’re talking about granting duty-free wouldn’t cover the hyperscalers alone, let alone the rest of the industry. Those are chips we physically can’t buy here, because the capacity doesn’t exist yet.”

Who Pays First if the Carve-Outs Die?

The loudest budgets on this file are not foundries. They are the U.S. designers and cloud firms that already cannot get enough wafers. A Center for a New American Security report found Microsoft, Alphabet, Amazon, Meta, and Oracle plan almost $700 billion in 2026 capital spending, most of it for AI plant. CSIS estimates server chips are at least 45 percent of modern data-center capital cost.

Jonathan McHale, digital policy chief at the Computer and Communications Industry Association, whose members include Amazon, Google, and Meta, has been carrying that point into the talks.

This data center buildout, in scale and dollars, has been compared to building the transcontinental railroad. Anytime you add to the cost and decrease predictability you make it more difficult to invest, and you are putting that in jeopardy.

Jonathan McHale, digital policy chief, Computer and Communications Industry Association

Nvidia and AMD design in the United States and still depend on overseas plants to make the dies. A duty on those imports lands on them before it lands on TSMC’s P&L. Apple faces a different cut: rivals abroad can buy the same chips without the U.S. tax, while a wider net on laptops and consoles would show up on U.S. shelves. Lobbyists also warn that allied suppliers, squeezed at the American dock, will look harder at China for orders.

White House spokesman Kush Desai said reshoring semiconductor manufacturing is a top priority for the president, “whose policies have already secured hundreds of billions of dollars of investments in this key sector.” The Commerce Department did not answer questions about the next round. One tech official from a major industry group, who also served in the first Trump administration, was blunter about the collision with the AI race: “This may be the single dumbest way imaginable to pursue American dominance in AI. It’s like kneecapping yourself at the starting line.”

THE GAP THE QUOTA HAS TO CROSS

• U.S. output share: The January proclamation says the United States fully manufactures only about 10 percent of the chips it requires.
• Arizona ceiling: TSMC still projects only about 30 percent of its 2-nanometer and more advanced capacity in Arizona when every announced plant is up.
• Duty-free math: The live Taiwan formula is 2.5 times planned capacity during construction, then 1.5 times U.S. output.
• Buyer spend: Five U.S. tech firms are aiming at almost $700 billion of 2026 capital spending, most of it AI infrastructure that still needs imported accelerators.

Every conversation with Commerce, that same tech official said, is some version of please tread carefully, because you do not understand how much demand is coming or how little capacity exists to meet it. The industry is not arguing against U.S. plants. “The data center industry would love to have a strong domestic semiconductor manufacturing industry, and we’re here to support that on-shoring,” one person in the sector said. “But the volume and scale of manufacturing of advanced node chips and memory chips is just not where it needs to be.”

The January Order Already Mapped Broader Duties

Phase 2 is being briefed as a new shock. The January 14 proclamation already described it. After a Section 232 investigation, Commerce found that chip, tool, and derivative imports threaten national security. The secretary recommended a two-phase plan: a narrow 25 percent duty now, then, after talks with trading partners, broader tariffs “at a rate of duty that is significant,” paired with a tariff offset for companies that invest in U.S. production and parts of the supply chain.

The White House fact sheet said the president may, in the near future, impose broader tariffs on semiconductors and their derivative products, with that offset program attached. The proclamation itself states that the United States fully manufactures only about 10 percent of the chips it requires, while it consumes roughly one quarter of the world’s semiconductors. That dependence, the secretary found, is the security problem.

So the irony is baked into the paperwork. Phase 1 taxed a thin slice of accelerators and then excused data centers, because those imports were said to help the U.S. tech stack. Phase 2, as now described in private, would pull that shield back and extend the duty to the machines those chips sit in, using the same investment-for-quota logic Taipei already signed. Michael Sobolik, a senior fellow at the Hudson Institute and a former aide to Sen. Ted Cruz, called getting supply chains de-risked one of the central geopolitical questions of our lifetime, then added that building a lot of these chips in the United States at scale is going to be really expensive, because there are a lot of cheaper places in the world to do it.

Tariffs Do Not Hire Technicians or Pour Concrete

Sujai Shivakumar, an economist who directs the Renewing American Innovation program at CSIS, has been making the unglamorous point that a duty can move a relative price and still leave the factory unbuilt. Higher tariffs, he said, do not produce more skilled technicians, shorten permitting timelines, expand reliable power and water, or create qualified suppliers. Arizona’s own pitch lists more than 70 semiconductor expansions since 2020, worth over $314 billion, which is real concrete. It is also still a minority of the wafers the AI halls will buy this year and next.

WHAT WE KNOW

• Phase 1: A 25 percent duty on a narrow set of advanced chips has been in force since January 15, 2026, with a long list of U.S. end-use carve-outs.
• The quota: Taiwanese firms that build U.S. plants can import 2.5 times planned capacity duty-free during construction and 1.5 times U.S. output after.
• The ceiling: TSMC’s $265 billion Arizona plan still places only about 30 percent of its most advanced capacity in the United States at full build-out.

WHAT IS UNCONFIRMED

• The rate: Commerce has not settled the Phase 2 percentage, the phase-in clock, or whether country quotas will apply.
• The carve-outs: Officials have signaled the January exemptions may die, but no new proclamation has withdrawn them.
• The product list: Servers, laptops, and consoles are in the talks; they are not yet in the tariff schedule.

The people in the talks still say the framework can change in the coming weeks or months. Until it does, the only working model is the one Lutnick already signed with Taiwan, and that model rations duty-free chips against plants whose advanced output, even on paper, stays mostly overseas.