AINEWS
An administration racing to win on AI is weighing a semiconductor tariff its own industry says could slow the data center build out the race depends on.
The Trump administration is weighing a new round of semiconductor tariffs that would extend well beyond raw chips to the products that contain them: laptops, gaming consoles, and the servers that power AI data centers. The proposal, reported by POLITICO and re-reported by LiveMint citing eight people familiar with the discussions, is still fluid and could change in the coming weeks, but the early shape puts the administration on a collision course with its own AI ambitions.
Tech companies are warning that higher costs on imported chips, and on the devices built around them, could delay or cancel planned data-center projects at the moment US demand for AI capacity is surging. The industry position, also carried by Yahoo Finance's policy team, calls the policy a tax on the very build-out the administration has identified as a national priority. The collision is sharp because the United States remains heavily dependent on Asian chip suppliers, including Malaysia, South Korea, and Taiwan. Taiwan alone produces more than 90% of the world's most advanced semiconductors, according to the LiveMint report.
The policy design the administration is gravitating toward, as described in the same reporting, is a relief-for-investment structure championed by Commerce Secretary Howard Lutnick. Under the approach, foreign chipmakers would see lower tariff exposure in exchange for commitments to build or expand fabrication capacity on US soil. The proposal also envisions phasing the duties in over time rather than imposing them all at once, a sequencing choice that would give companies time to reshore but would not insulate them from higher near-term input costs on the servers their customers have already ordered.
The administration imposed a 25% tariff on February 9, 2026, on certain advanced semiconductors under Section 232, the trade-law authority used when imports are deemed a national-security threat, according to a White & Case alert and CNBC's earlier tariff coverage. That earlier action hit the chip itself. The new proposal would broaden the scope from the chip to finished goods, and would tie the relief mechanism to factory investment rather than country of origin.
The relief-for-investment swap is the mechanism that decides who wins. It rewards capacity built in the United States regardless of where the parent company is headquartered, the part of the design the administration will defend as pro-reshoring. It is also the part most likely to be contested by trading partners, since it conditions market access on capital commitments that look like industrial policy by another name. The tariff would function less as a tax on imports and more as a bidding system for US fabs.
The data-center warning lands on the AI build-out, where the US is already short of high-end chips. LiveMint and Yahoo both describe tight supply as a binding constraint on new training and inference capacity. Adding a tariff on top of that constraint, even a phased one, would push the cost of the next round of servers higher and stretch delivery timelines, according to the industry's reported warnings. The Lutnick structure is meant to address that pressure by accelerating domestic fab construction, but fabs take years to come online and the demand surge is measured in quarters.
Three policy decisions in the next few weeks will determine whether the trade-off closes. The first is the scope of downstream products the final proposal covers, and whether data-center servers land in or out of the duty. The second is the size of the relief offered for each dollar of US fab investment. The third is the phasing schedule. Each one shifts the math for the data-center projects the industry says are most exposed.