By Reva Goujon and Ben Reynolds

About the authors: Reva Goujon is a director at Rhodium Group. Ben Reynolds is a research analyst with Rhodium's China corporate advisory team.

President Donald Trump's second term has seen several staggering deal announcements to advance domestic investment in artificial intelligence. But his push for an industrial revolution in AI is in significant tension with his plans for wide-ranging tariffs on semiconductors. In particular, a handful of companies can make or break his ambitions, and if he wants them to deliver, he'll need to get creative with his tariff plans.

At Trump's order, the U.S. Commerce Department is moving to complete a Section 232 investigation on the national security impact of semiconductor imports. Tariffs stemming from this order would target not only chips directly imported to the U.S. but also products containing chips, including smartphones, servers, and laptops. The tariffs would also target "foreign monopoly exposure" in manufacturing tools used in chip fabrication, which would include tools essential to cutting-edge semiconductor manufacturing like Netherlands-based ASML's advanced lithography machines and Tokyo Electron's coater/developer tools.

Trump's strategy assumes that high tariffs will drive companies to build facilities in the U.S. He is also using the threat of high "reciprocal" tariffs to extract investment deals from abroad. This approach faces practical challenges, however. Foreign governments have limited power to force companies to the negotiating table, and companies will only commit to investments if the economics make sense. High tariffs on the critical inputs to build advanced fabs and AI data centers in the U.S. would destroy the credibility of any investment pledge made on a geopolitical pretense.

Tariffs' Impact on AI Infrastructure

The final semiconductor tariff rate hasn't been set, but if it is anything like other sectoral tariffs (25%), it would increase the cost of each major advanced fabrication project currently under way in the U.S. by roughly $4 billion on average. These projects are supported by some $3.9-6.6 billion in grants from the 2022 Chips Act, which were intended to offset higher costs of building in the U.S. Tariffs on semiconductor manufacturing equipment would effectively negate Chips Act funding.

The impact on AI data centers could be even more dramatic. The tariff-free cost of equipping a hyperscale data center with servers containing 100,000 Nvidia H100s costs $3-$3.5 billion based on current pricing. A 25% tariff would raise that cost by at least $750 million.

Nvidia intends to shift production of some of its graphics processing units to Taiwan Semiconductor Manufacturing's Company new facility in Arizona. That would partially mitigate the tariff impact, but TSMC's U.S. capacity is limited. The U.S. is also completely dependent on foreign sources for high-bandwidth memory, another essential AI processor component. Even in the most optimistic reshoring scenario, imports will remain a key enabler of U.S. AI infrastructure buildouts over the next few years.

TSMC has pledged to invest $100 billion in U.S. facilities, while Nvidia and its partners are planning $500 billion worth of AI infrastructure. The Trump team frequently cites these investments as evidence that its tariff plans are working to bring manufacturing to the U.S. However, whether these projects actually break ground within the next four years depends on how Trump reconciles his tariff strategy with his AI ambitions.

The Chosen Few

The companies that hold the keys to Trump's AI vision include Nvidia as the premier AI chip designer; TSMC as the leading logic chip manufacturer; Micron, Samsung, and SK Hynix as the biggest suppliers of high-bandwidth memory; key equipment manufacturers like ASML and Tokyo Electron; hyperscalers like AWS, Microsoft Azure, and Google Cloud; electronics giants like Apple; and manufacturers and assemblers like Taiwan-based Wistron and Foxconn.

This cluster of companies has immense power to pull in demand and steer investment. Each one's decisions can have powerful force-multiplying effects on the others -- and their host governments. For example, Taiwan, Japan, and South Korea are all critical to semiconductor and electronics supply chains and all depend heavily on the U.S. for defense support. The Trump administration is using the defense relationship as leverage to draw in tech investment. But there remains a gaping hole in that strategy. All these companies face similar questions: What are the added project costs from tariffs, what incentives will offset those costs, and how will critical labor gaps be filled? Those questions are complicated by the possibility of a tariff-induced economic slowdown that stunts AI demand and disrupts billions of capital expenditures in planned AI infrastructure buildouts. With billions in potential added costs from Section 232 duties, Trump will need to find creative offsets.

Reshaping Incentives

The Chips Act provides for $39 billion in grants and 25% investment tax credits to reshore chip manufacturing, plus $13 billion for research and workforce training. Trump has condemned this approach, arguing he could secure more investment through negotiations without taxpayer money. This stance, along with personnel reductions at the Chips for America office, has raised concerns about whether the U.S. administration intends to claw back some funding.

Commerce Secretary Howard Lutnick has indicated that while the administration doesn't favor direct grants, it may consider expanding the 25% tax credit. Companies may welcome that approach, since they generally prefer tax credits over grants. Tax credits provide flexibility over complex construction timelines, are easier to claim, and less susceptible to politicization than milestone-based grants.

Two bills in Congress -- the Building Advanced Semiconductors Investment Credit (BASIC) Act and the Semiconductor Technology Advancement and Research $(STAR.UK)$ Act -- would enhance those credits. Expanding the Chips tax credit from 25% to 35%, as the BASIC Act proposes, would free up an average of $4 billion in incentives for each of the four major U.S. advanced chipmaking projects currently under construction, enough to offset the additional $3.8 billion in average costs associated with a 25% tariff on semiconductor tools. The STAR Act would extend the expiration date for the tax credit from 2026 to 2031, providing the certainty firms need to move forward with more ambitious long-term plans to expand on their initial investments.

Trump will need to be flexible in tariff implementation if he wants to make real progress in advancing semiconductor and AI projects in the U.S. Carve-outs for critical inputs, reasonable phase-in timelines, and duty deferrals through free trade zones will be essential.

Foreign trade zones allow companies to defer duty payments on imports, particularly valuable during tariff volatility. Notably, Trump's auto tariffs took a hard line and overruled potential foreign trade zone benefits. The administration has also made ad hoc adjustments after initial tariff announcements, introducing exemptions and offsets when faced with industry pushback.

The Section 232 semiconductor tariffs need not follow the same chaotic path. The president has tech leaders in his orbit who understand what's required for major AI infrastructure investments. They will understand the necessity of tying incentives such as tax rebates and rational exemptions on duties with credible investment deals that boost manufacturing in the U.S.

The stakes are high for a capital-intensive industry at the epicenter of U.S.-China competition. If Trump wants to be remembered for accelerating the AI industrial revolution, he may yet see the virtue in enabling -- rather than hindering -- the investments that could cement that legacy.

Guest commentaries like this one are written by authors outside the Barron's newsroom. They reflect the perspective and opinions of the authors. Submit feedback and commentary pitches to ideas@barrons.com .

This content was created by Barron's, which is operated by Dow Jones & Co. Barron's is published independently from Dow Jones Newswires and The Wall Street Journal.

(END) Dow Jones Newswires

May 13, 2025 08:15 ET (12:15 GMT)

Copyright (c) 2025 Dow Jones & Company, Inc.