Huawei is China's designated national champion for AI compute, and the US export control regime is the reason why. The US currently holds over 70% of the world's deployed FLOPs. To protect that lead, Washington has blocked Nvidia GPU sales to China and restricted access to advanced fabrication equipment. The policy worked well enough that DeepSeek's next-generation model training was delayed by reliance on Huawei chips, and most of DeepSeek's inference still runs on western hardware.
The tradeoffs are now visible on both sides. China retaliated by cutting rare earth minerals and magnets, forcing Commerce Secretary Howard Lutnick to resume Nvidia H20 sales to China in exchange for restarting those supply chains. Meanwhile Huawei is ramping Ascend chip production into the millions of units this year, with HBM supply projected to become the binding constraint in 2026. The full production ramp breakdown, including die bank strategies and the continued role of TSMC, is what this piece actually documents.
This is not a stable state. China built its domestic internet industry by expelling Google behind the Great Firewall. It is now applying the same logic to silicon. The original article covers Huawei's vertical integration across fabs, tooling, and chip design, and why the HBM ceiling matters more than any single policy decision. Read it for the supply chain specifics, not the geopolitics.
[READ ORIGINAL →]