Commerce Secretary Howard Lutnick is pitching a new semiconductor tariff with a duty free carve out for companies that build U.S. chip factories. Those are the same chips the AI race runs on.
The Trump administration is weighing a new round of tariffs on imported semiconductors, the same chips the White House has put at the center of its plan to "win the AI race" against China. Under the mechanism reported by Politico, foreign chipmakers could escape duties by committing to U.S. manufacturing investment, with the size of the carve-out scaled to how much a company spends building chip factories on American soil.
Commerce Secretary Howard Lutnick has been pushing the arrangement inside the administration, according to anonymous sources cited in the reporting. The pitch is essentially industrial policy inside a tariff regime: keep the duties on, but tie exemptions to onshore chipmaking capacity.
Building a fabrication plant, known in the industry as a fab, from groundbreaking to high-volume production takes three to five years on the optimistic end. Tariff incidence is immediate. A duty on imported semiconductors hits the price of every laptop, server, and AI accelerator bought while the U.S. fabs are still under construction.
The Computer & Communications Industry Association, a trade group representing large platform and hardware buyers, calculated in 2025 that extending the tariff regime to data centers would cost the U.S. economy roughly $90 billion a year. That is an industry-association cost projection, not an independent government or academic number, but it is the figure the affected buyers are using to push back.
The current tariff regime any new round would extend is White House Proclamation 11002, issued in January 2026, which adjusted duties on semiconductors, the lithography machines that make them, and their derivative products. The underlying authority is a Section 232 national security investigation launched in 2025 and finalized into the Federal Register rule (2026-01052). Any new tariffs would layer on top of that base.
The contradiction is structural, not just rhetorical. Trump has made winning the AI race a stated top priority. In practice, that has meant clearing obstacles to data center construction, brokering multibillion-dollar deals with TSMC, and taking a government stake in Intel. TSMC, which holds roughly 73% of the global semiconductor foundry market and supplies Nvidia's processors, is the linchpin of those investment pledges.
Meanwhile, the AI buildout is already tightening the global chip supply. AI data centers are pulling memory and advanced packaging capacity that used to serve laptops and consumer electronics, and laptop prices are already climbing as a result. A tariff on the same chips the AI buildout is consuming would add a second upward push on the same prices.
The Semiconductor Industry Association, which represents U.S. chipmakers, issued a statement warning that the tariff regime was raising costs and slowing investment, and CCIA organized a joint association letter with other trade groups on the same theme.
The Lutnick proposal's defenders say the invest-to-evade trade is the only way to make the tariff regime politically sustainable: the duties create the incentive, the carve-outs let compliant companies breathe, and the U.S. ends up with more domestic fabs in the medium term. Its critics, including some of the same trade associations, say the timing mismatch is the whole point. The U.S. cannot tariff its way to fabs that take half a decade to come online without first taxing the AI buildout it has spent two years subsidizing.
What would have to be true for the plan to function: a U.S. fab pipeline concrete enough to credit, on a schedule that overlaps the tariff's first effective years. That schedule has not been published.