McKinsey Names Electronics America’s Most Dangerous Import — And It’s the Entire AI Supply Chain

McKinsey’s May 2026 manufacturing report reveals a single, uncomfortable truth: the $900 billion import category that powers the AI boom is also the most trade-exposed sector in the entire US economy.

What Happened

The McKinsey Global Institute’s May 2026 report, Ramping Up Manufacturing in America, mapped every major US import sector by two axes: import volume and trade exposure. Electronics sat alone at the worst intersection of both. At $900 billion in 2025 imports, it dwarfs every other category — Transportation equipment at $450B, Chemicals at $410B, Metals at $300B, and Machinery at $230B don’t come close individually, let alone in combined vulnerability.

MGI defined “trade-exposed” with precision: a sector qualifies only when it scores on all three dimensions simultaneously — the inputs are critical (no easy substitutes), the supplier base is concentrated (few countries or firms dominate), and those suppliers are geopolitically distant (relationships subject to export controls, sanctions, or alliance fragility). Electronics hits all three harder than any other sector in the US import basket.

The structural implication is not abstract. Chips, memory, servers, and networking hardware — the physical substrate of every AI data center being built right now — live inside that $900B line item. The AI supercycle is, at its foundation, a massive bet placed on the most geopolitically fragile supply chain in American trade.

The Structural Read

The McKinsey data only makes sense when you hold it next to a parallel reality: while the US imports $900B of electronics from concentrated, distant suppliers, China has been doing the opposite — deploying productive capital domestically to build the factories, fabs, and supply chains that produce those same goods. That asymmetry is not an accident. It is a decade of deliberate industrial strategy meeting a decade of US consumption-over-production preference.

The AI supercycle has made this asymmetry dangerous rather than merely inconvenient. When AI was a software story, hardware dependency was a manageable input cost. Now that AI is a compute infrastructure story — measured in GPU clusters, HBM memory stacks, and custom silicon — every major model training run, every inference deployment, every hyperscaler capacity expansion traces directly back to that $900B import line. The bottleneck is physical, not algorithmic.

This is also why the chip war, BIS export controls, TSMC’s Arizona buildout, and the DRAM antitrust spotlight (Samsung, SK Hynix, Micron) all feel like separate stories but are actually one story. They are different governments and firms responding to the same structural chokepoint McKinsey has now quantified in dollar terms. The reshoring push is not industrial nostalgia — it is a logical response to the discovery that the most important supply chain in the AI economy has all three of the worst possible dependency characteristics at once.

Where This Lands on the AI Stack

The Map of AI framework — which tracks 200+ companies across 9 layers of the AI stack — makes the vulnerability legible at a layer-by-layer level. The trade exposure is not evenly distributed across the stack. It is maximally concentrated at the bottom two layers, precisely where the US has the least domestic production capacity.