The AI chip geopolitics update: 2026 edition

I update this analysis every year. The map changes. The fundamental problem doesn’t.

The country-by-country picture

Taiwan: Still makes approximately 90% of the world’s most advanced chips (sub-5nm). TSMC is dominant. They’re also 100 miles from a geopolitical flashpoint. The concentration risk hasn’t changed. The concentration itself hasn’t changed.

United States: One TSMC fab in Arizona producing 4nm chips. One more under construction for 3nm. Intel is struggling. Their 18A process is behind schedule. The Ohio fab is delayed. The CHIPS Act money is flowing, but fabs take 3-5 years to build and even longer to reach competitive yields. The US is better positioned than two years ago. It’s still far from self-sufficient.

South Korea: Samsung is at 2nm with improving yields. Their GAA nanosheet architecture is competitive. SK Hynix dominates high-bandwidth memory for AI. Korea is the strong second player, and the gap with TSMC is narrowing.

China: SMIC is producing at 7nm without EUV, using multi-patterning DUV lithography. It’s slower, lower yield, and more expensive, but it works. China is building massive capacity at older nodes (28nm, 14nm) that serve the majority of chip demand (cars, IoT, appliances). The export controls pushed China toward self-sufficiency faster, though at a lower performance tier.

Japan: Rapidus is building a 2nm fab in Hokkaido with IBM technology licensing. Japan hasn’t manufactured leading-edge chips in decades. This is a moonshot. The timeline is aggressive. The ambition is real. Whether it succeeds will define Japan’s role in semiconductors for a generation.

Europe: ASML remains the monopoly lithography supplier. But Europe has almost no advanced chip manufacturing. The European Chips Act is smaller than the US version. No European fab is currently on a path to sub-5nm production.

What’s better

The US has a fab producing advanced chips. Two years ago it had zero.

Samsung is closer to TSMC at the leading edge. Competition improves the supply chain.

Japan is making a serious attempt to re-enter advanced manufacturing.

What’s worse

Intel’s foundry business is losing money and losing schedule. The most important American chip manufacturer might not survive as a manufacturer.

China is more self-sufficient at older nodes, which means export controls have diminishing leverage.

The number of operational High-NA EUV machines is in the single digits. All from ASML. The chokepoint hasn’t widened.

The earthquake test

Here’s my annual exercise. If a magnitude 7.0 earthquake hit Hsinchu, Taiwan (where TSMC’s main fabs are located), what happens?

The world loses access to approximately 90% of its most advanced chips. For months at minimum. Possibly longer. Apple can’t make iPhones. NVIDIA can’t make AI GPUs. Qualcomm can’t make mobile processors. The auto industry (already chip-constrained) collapses.

The economic damage would be in the trillions. Not billions. Trillions.

The Arizona fab helps, but one fab making 4nm chips can’t replace the output of twelve TSMC fabs making everything from 3nm to 5nm to 7nm.

This risk is real. It’s quantifiable. And it’s the single biggest infrastructure vulnerability in the global economy. It was true when I wrote this analysis last year. It’s still true.

The map is being redrawn. But the redrawing isn’t fast enough.

Related thinking: