Author
American imports related to the artificial intelligence boom have more than doubled since 2023 while non-AI imports have fallen. This hunger for AI investment has increased the U.S. trade deficit despite the highest tariffs in a century. AI-related exports from the U.S. have also surged, though not as much, according to new (AI-assisted) research from Minneapolis Fed Monetary Advisor Michael Waugh (Minneapolis Fed Staff Report 684, “Trade in AI-Related Products”).
By tapping a large language model (LLM) to analyze data for more than 18,000 products, Waugh pushes beyond obvious AI inputs, like computer hardware, to capture the broad range of goods related to the build-out of data centers and other infrastructure. It’s not just microchips and circuit boards: A sharp rise in categories such as electric power, cooling HVAC, and telecommunications clearly accompanies the recent investment in AI.
Every imported good has a Harmonized System 10-digit (HS10) product code. Waugh sorts the universe of HS10 codes according to AI relevance (high, medium, low) by prompting the LLM to cross reference them with industry information from the North American Industry Classification System (NAICS). The resulting 645 codes with high AI relevance capture not just microprocessors but general-use products like refined copper cathodes (with a 150 percent increase in import dollar volume since 2023). “Focusing on a narrow set of semiconductor codes,” Waugh writes, “would miss these products entirely.”
As of January 2026, the analysis finds imports of AI-relevant goods were 111 percent higher in nominal dollars than the monthly average in 2023 (see figure). Imports with low AI relevance were down 14 percent. Prior to 2024, these high and low AI-relevant bins displayed nearly identical trends.
Another way to quantify the change: For all of 2025, total dollar imports of AI-relevant products ($379 billion) were 72.6 percent higher than in 2023. Imports of products with low AI relevance rose just 2.5 percent.
This growth occurred alongside historic increases in U.S. import tariffs during 2025. Using methods from his tariff-related research, Waugh calculates that AI-relevant goods faced an effective tariff rate of just 4.5 percent, versus 12.1 percent for non-AI goods. Tariff exemptions—principally one for consumer electronics—covered about 69 percent of AI-relevant imports, a situation that remains largely unchanged after February’s Supreme Court rule invalidated some tariffs.
AI trade flows both ways. U.S. exports related to AI were 35 percent higher in 2025 than in 2023. But the rise in imports was much larger. In a counterfactual exercise, Waugh finds that without these AI effects on trade, the U.S. trade deficit would have been 16 percent smaller in 2025.
The analysis unsurprisingly finds that Taiwan, the world’s dominant producer of advanced semiconductor chips, is a major source of AI-relevant imports. But under Waugh’s broader umbrella of AI-related goods, Mexico is equally important, with Mexico and Taiwan each supplying about a quarter of U.S. imports. Mexico is a major source for electrical, cooling, and networking products.
Mexico is also a major destination for America’s AI-related exports. Some of this traffic, Waugh believes, represents supply chains that cross the southern border multiple times. AI-related trade also helps explain why U.S. trade with Mexico remained robust in the face of tariffs while imports from Canada dropped.
AI-related imports from mainland China were never large, but have fallen since 2023 and are now below 10 percent, less than Vietnam or Thailand. Waugh notes that Chinese imports—AI-related and otherwise—faced an additional 20 percent tariff (based on allegations of supplying chemicals used to make the drug fentanyl).
To test the strength of the findings, Waugh tests variations on his LLM product-classification prompt as well as a coarser, non-LLM matching technique, yielding similar results. For reference, he also asked the model to provide natural language explanations for each of its 18,364 classification decisions. While the technique can likely be refined and further validated, the early results display the extent to which AI is steering the overall course of U.S. trade—perhaps, Waugh writes, “even more important than dramatic changes in U.S. trade policy.”
Read the Minneapolis Fed staff report: “Trade in AI-Related Products”
Jeff Horwich is the senior economics writer for the Minneapolis Fed. He has been an economic journalist with public radio, commissioned examiner for the Consumer Financial Protection Bureau, and director of policy and communications for the Minneapolis Public Housing Authority. He received his master’s degree in applied economics from the University of Minnesota.
