The Physical Stack of AI · Fab and packaging supply chain
You can describe the four-year arc of US chip export controls — from the October 2022 baseline to the case-by-case sales regime of January 2026 — and explain why both Washington and Beijing now treat advanced semiconductors as a national-security file.
In October 2022 the US Bureau of Industry and Security published a set of export rules that — for the first time — applied to AI accelerators rather than weapons. The intent was to slow China's progress toward training frontier AI models. The mechanism was a ban on selling Nvidia's A100 and H100 to mainland China, paired with restrictions on the lithography and tools needed to fabricate them. Three escalations followed: the October 2023 update closed the H800/A800 workaround Nvidia had built; the December 2024 package extended controls to HBM and semiconductor-manufacturing equipment; and Biden's final-week AI Diffusion Rule of January 15, 2025 layered a worldwide three-tier licensing framework on top.
Then the politics shifted. The Trump Commerce Department rescinded the Diffusion Rule on May 13, 2025 — two days before it would have taken effect — calling it overreach. In December 2025 the administration approved case-by-case sales of cut-down H200 and MI325X parts to mainland China; a BIS final rule of January 13, 2026 formalized the new "case-by-case review, not presumption of denial" posture. The result is best taught as a bifurcated regime: some H200- and MI325X-class exports can be licensed case by case, while training-class compute, HBM, and the manufacturing inputs behind domestic substitutes remain tightly controlled.
Chapter contains 4 lessons.