China's AI Chips Backfilling Potential

This is a cross-post of my X thread.

Some claim that if the US controls AI chips, countries will immediately turn to China to backfill.

Critics rightly identify this as a crucial consideration—but I don't think it's an immediate and strong threat. Here's why:

1. China's ability to "backfill" is significantly constrained.
a) The validated end user requires cutting business ties with Chinese companies and not using their hardware. You can't be in both camps.

b) Existing export controls limit both quality and quantity of chips China can produce.
I've not seen a single cluster with Chinese AI chips outside of China. @KonstantinPilz tracked the 500 most performant AI supercomputers globally:

I've previously shared that I expect Huawei to get more chips than ever this year (via TSMC):

I expect this is mostly for domestic demand, but if I'd be in charge I'd build a cluster elsewhere just to rattle confidence. Just so you know.

The US maintains technological leadership across the entire AI stack. With a ~4 year lead in frontier AI chips, even Chinese users have been slow to adopt domestic alternatives due to software limitations and performance gaps.

Countries make strategic technology partnerships based on broader considerations beyond AI. Military cooperation, security guarantees, and existing relationships typically outweigh marginal differences in chip performance.

Getting the balance between security controls and market access right is crucial. While China's ecosystem currently poses limited competition in scale, maintaining US leadership requires careful calibration of restrictions.

Some of the points from my paper "Understanding the Artificial Intelligence Diffusion Framework."