Based on McKinsey Global Institute (reporting via SCMP).
AI-related goods drove roughly one-third of all global trade growth in 2025, the US captured +66% of that surge, and the supply chain running through Taiwan, South Korea, and ASEAN is now the physical backbone of the AI supercycle.
What Happened
In its March 2026 update, Geopolitics and the Geometry of Global Trade, the McKinsey Global Institute published a number that reframes the entire conversation about AI’s economic weight: AI-related goods accounted for approximately one-third of all global trade growth in 2025. Global AI-related trade expanded roughly 37% in the 2024–25 window. Every other category — advanced manufacturing excluding AI, resources, basic manufacturing — grew in the low single digits or contracted outright.
The US figures are the sharpest edge of the data. American AI-related goods trade grew approximately 66% — more than four times the EU’s 22% and more than four times China’s 16%. That gap isn’t a policy accident; it’s a direct consequence of the data-center buildout, which placed roughly half of all new global data-center construction on US soil in the same period. Import demand for semiconductors, servers, and networking gear followed. Meanwhile, US advanced manufacturing (ex-AI) fell approximately 1% and basic manufacturing fell approximately 2% — capital and import capacity are being actively reallocated toward compute, not layered on top of existing industrial activity.
The mechanism McKinsey identifies is supply-chain geography. The data-center boom runs on hardware that flows through a tightly integrated corridor: Taiwan (leading-edge logic), South Korea (memory, advanced packaging), and ASEAN (assembly, substrates, networking components). That is precisely why Taiwan recently overtook China as a source of US imports — a structural shift in the trade map driven not by tariff arbitrage but by where the chips are actually made.
The key insight: When one category drives a third of all global trade growth while the rest stagnate or shrink, global commerce isn’t diversifying — it’s converging around a single physical bottleneck: compute. The money that Goldman Sachs mapped onto a $1 trillion capex curve becomes chips and servers at the fab level, and chips and servers become trade flows at the port level. These are the same phenomenon, measured at different altitudes.
The Structural Read
The McKinsey data is not an AI story dressed up as a trade story. It is a trade story that exposes the physical architecture of the AI supercycle — and it has three structural layers worth separating.
First: trade now has an AI signature. The divergence between AI-related goods (+37% global) and everything else (low single digits or negative) is the dockside version of a pattern visible across the capital stack. NVIDIA’s data-center revenue grew approximately 1,300x over the cycle. Hyperscaler capex is converging on a $1 trillion annual run rate. Goldman Sachs, Morgan Stanley, and now McKinsey are each measuring the same underlying force at different points in the value chain. At the port level, it shows up as this: one category dominates, and the rest recede.
Second: the US is the demand epicenter, and it is reallocating, not adding. A +66% surge in AI-goods trade alongside -1% and -2% declines in advanced and basic manufacturing is not a coincidence. It is capital reallocation made visible in import data. The US economy is concentrating its industrial capacity toward compute — and the trade flows reflect that priority shift in real time.
Third: geopolitics reshapes routes, not volume. This is the most counterintuitive finding. The AI hardware surge happened in the middle of escalating trade tensions, export control tightening, and deliberate decoupling from China. The result was not a slowdown — it was a rerouting. China’s AI-goods trade grew a modest 16%; Taiwan, South Korea, and ASEAN absorbed the demand that might otherwise have flowed through Chinese supply chains. The AI buildout is powerful enough to override geopolitical friction. Decoupling changes the map’s geometry; it does not shrink the map’s volume.
Map of AI — Physical Layer
The Trade Map and the Compute Map Are Now the Same Map
The Map of AI framework places compute infrastructure at the foundation of the entire stack — every model, every application, every vertical built on top runs through this layer. McKinsey’s trade data is the first time that physical reality has appeared at scale in macroeconomic statistics. The Taiwan–Korea–ASEAN corridor is not a supply chain for a single company or product; it is the substrate of the AI economy. Control of that corridor — through fabs, through export rules, through port capacity — is the new industrial policy frontier.
McKinsey Global Institute — Geopolitics and the Geometry of Global Trade, 2026
“AI-related goods accounted for approximately one-third of all global trade growth in 2025, with AI hardware and the data-center buildout the dominant engine — while every other major trade category grew in low single digits or contracted.”
