The AI boom is singlehandedly carrying the U.S. import market—and adding $200 billion to the trade deficit, Fed study finds

AI-related products accounted for 23% of all U.S. imports last year, with $265 billion of AI imports versus $71 billion of AI-related exports, underscoring how the AI buildout is driving trade flows and widening the goods trade deficit. The article says AI-related imports have risen 73% since 2023, while non-AI imports were up just 3%, and effective tariff rates on AI goods were only 4.5% at the end of 2025 versus 12.1% for non-AI goods due to broad exemptions. The piece highlights that domestic manufacturing still cannot meet data center and semiconductor demand.

Analysis

The market is underestimating how much the AI capex cycle has become a de facto fiscal stimulus for selected industrial and hardware supply chains. The key second-order effect is that tariff policy is not suppressing imports uniformly; it is reshaping them toward exempted, high-value AI inputs, which means the trade deficit is being supported by a narrow set of categories rather than broad-based demand. That concentration makes the macro impulse fragile: if data center orders normalize even modestly, headline import growth could roll over quickly because the rest of the basket is already soft.

TSM remains structurally better positioned than INTC because the bottleneck is leading-edge process capability, not domestic political optionality. The article reinforces that U.S. onshoring does not solve near-term AI compute demand, and any reshoring of advanced chips is a multi-year capacity story constrained by labor, permits, and yield risk. INTC’s domestic fabrication narrative is therefore more of a policy proxy than a direct earnings catalyst; the company can win incentives, but not necessarily share, if hyperscaler demand keeps prioritizing proven supply.

The contrarian miss is that low effective tariffs on AI inputs may actually prolong dependence on Asia by protecting the buildout economics. If exemptions persist, they function as a subsidy to hyperscaler capex, which is mildly bullish for semiconductor equipment and foundry capacity but bearish for policy-sensitive reindustrialization plays. The risk to this trade is a sudden exemption rollback or a capex pause from the largest cloud buyers, either of which would hit suppliers over a 1-3 month horizon before showing up in macro data.

Short term, the setup favors relative-value rather than outright directional trades: the AI import complex can remain strong even if the broader economy slows, but the margin capture will stay abroad unless domestic manufacturing bottlenecks clear. That argues for leaning into beneficiaries with actual supply-chain share, while fading names whose upside depends on a faster U.S. fabrication ramp than the operating reality allows.

Key Decisions for Investors

• Long TSM / short INTC for a 3-6 month relative-value trade: TSM owns the scarce leading-edge supply, while INTC’s upside is still mostly policy optionality. Risk: a materially larger U.S. subsidy package or accelerated Intel execution could compress the spread.
• Buy TSM call spreads 6-9 months out on any pullback: if AI capex remains the marginal driver of imports, foundry demand should stay supported even if non-AI manufacturing weakens. Use defined-risk structures because a capex pause would hit the premium leg quickly.
• Avoid chasing INTC on reshoring headlines; instead, trade it tactically around policy events only. Best expression is sell strength into announcements unless there is concrete evidence of backlog conversion or yield improvement.
• Pair long semiconductor infrastructure beneficiaries vs. domestic industrial laggards: e.g., long AMAT/LRCX vs. short XLI basket over the next 1-2 quarters. The thesis is that AI capex pulls through tools and services faster than it rescues broad manufacturing.
• Set a watchpoint on hyperscaler capex guides over the next earnings season; if the top four cloud providers slow spend growth by >10% sequentially, reduce exposure across the AI hardware complex immediately because the import impulse is concentrated and can reverse faster than consensus expects.