US mulls tariffing devices based on the number of chips used, estimated value (www.tomshardware.com)

The U.S. is considering new import tariffs that would target electronics by chip content — charging a percentage based on an estimated semiconductor value inside a device (a draft suggests ~25% for chip‑dense products, 15% for items from Japan/EU) — alongside a stricter proposal to require a 1:1 domestic chip production-to-import ratio or face penalties up to 100%. Both measures are still under internal review, but the mechanics would shift trade policy from product-level duties to chip‑aware pricing and could extend even to fab equipment currently exempted. For the AI/ML ecosystem this is material: high-density AI servers and GPUs would be among the costliest targets, potentially raising prices for training-inference infrastructure and slowing procurement. The plan raises major technical and operational headaches — estimating a chip’s “value in device” is nontrivial for multi-die packages, chiplets, or SoCs, and a unit‑count rule doesn’t account for huge value/complexity variation between tiny MCUs and advanced accelerators. Enforcement would require detailed tracing, harmonized definitions of “chip,” and new auditing systems across supply chains. Consequences could include accelerated onshoring investments (more U.S. fabs, higher capex and tooling costs), product redesigns to reduce chip counts or increase integration, and short‑term supply constraints and price inflation for AI hardware.