Vital Trades: AI goods +80% YoY

The real story here is not just “AI goods are up.” It’s that U.S. trade has split into two different economies. One is the AI buildout — servers, chips, storage, cooling gear, telecom equipment, copper, all the boring physical stuff that turns a model into a working data center. The other is basically the rest of the goods economy, which has been softer under tariffs and weaker demand. What changed this week is that Michael Waugh’s new Minneapolis Fed work started getting broader attention, and it gives the cleanest picture yet of how extreme that split has become. ### What counts as an “AI good” here? Not just GPUs. Waugh built a classification system across more than 18,000 U.S. trade product codes to capture goods used in building and running AI infrastructure. That means semiconductors, computers, and networking gear, but also power equipment, HVAC and cooling systems, and even refined copper cathodes. Basically, if it helps build a data center or feed one, it can land in the AI bucket. ### How big is the divergence? Big enough that the usual trade story breaks. In the Minneapolis Fed summary, AI-relevant imports in January 2026 were 111% above the average monthly level in 2023. Imports with low AI relevance were down 14% over the same span. The paper version puts the 2025 share in another useful way — AI-related products accounted for 23% of all U.S. imports in 2025, and those imports had grown 73% since 2023. ### Why didn’t tariffs crush this too? Because a lot of the most important tech categories got carved out. On April 11, 2025, the White House clarified that products under a list of HTS codes tied to semiconductors and electronics were excepted from the reciprocal tariffs, retroactive to April 5, 2025. The exemption covered categorically under the earlier IEEPA orders. So this was not “no tariffs.” It was selective relief for exactly the kinds of goods the AI boom needs most. ### Is this only about chips? No — and that’s the part people miss. If you only watch semiconductor import codes, you miss the second-order buildout. Data centers need switchgear, cooling, backup power, cabling, racks, telecom links, and industrial inputs. Waugh’s point is that the AI boom shows up across a wide physical supply chain, not just in one glamorous chip category. That’s why the import surge can stay large even when the headline tariff regime looks hostile. ### What does this do to the trade deficit? It keeps the deficit wider than tariffs alone would suggest. The Minneapolis Fed summary says the U.S. trade deficit would have been 16% smaller in 2025 without the effects of AI on trade. Fitch makes the same basic point from another angle — between February 2025 and February 2026, total basket weakened. ### Where are these goods coming from now? The sourcing map is shifting, not disappearing. Fitch says China’s share has fallen while Taiwan, Vietnam, and Mexico have gained. The Minneapolis Fed note also flags Mexico as a major source of AI-related imports and a destination for growing U.S. AI exports. So the system is rerouting around tariffs more than it is shrinking. ### Why does this matter beyond trade nerds? Because it tells you what the U.S. economy is actually prioritizing. Tariffs were supposed to suppress imports broadly. Instead, the AI buildout is so capital-hungry that it is overpowering that pressure in the categories that matter most to data centers. S&P Global saw the same pattern in February, when computers, accessories, and semiconductors accounted for $8 billion of a $14 billion jump in nominal goods imports. ### Bottom line? This isn’t just a weird chart. It’s a map of industrial policy in practice — punish broad goods trade, quietly protect the hardware stack behind AI, and watch capital flow there anyway. The result is an economy where “imports are down” and “AI imports are exploding” can both be true at the same time.