Author
Abstract
This paper documents facts about international trade in AI-related products. I develop a large language model (LLM) classification tool that maps HS10 codes in U.S. trade data to products used in the construction and operation of AI infrastructure. AI-related products account for 23 percent of U.S. imports in 2025, and imports of these products have grown by 73 percent since 2023. Over the same period, imports of non-AI-related products have grown by only 3 percent, with the divergence between the two categories beginning in early 2024. Mexico is a key market on both the import and export side, and together with Taiwan these two countries account for about half of all U.S. trade in AI-related products. Trade policy has treated these products lightly with product-level exemptions shielding much of AI-related imports from tariffs. Absent the AI boom, a simple accounting exercise suggests that the U.S. goods trade deficit would have been nearly $200 billion smaller in 2025.