THE CONTROL REGIME
Since October 2022, the US has restricted exports of advanced AI chips and the equipment to make them to China. The rule targets compute density — chips above specific performance-per-area thresholds need a license, which is presumed denied for Chinese buyers.
THE H200 AND ITS COUSINS
Nvidia built a sequence of China-specific chips — H800, H20, H200 variants — each engineered to sit just below whatever performance ceiling Washington set. Each tightening of the rule triggered a new SKU. The cat-and-mouse is the entire commercial relationship.
WHY CEOS ON AIR FORCE ONE
American presidents have flown executives to Beijing since Nixon — the practice signals that the visit is about commerce, not just diplomacy. Clinton took 24 CEOs in 1998; Obama brought a smaller cohort in 2009; Trump's 2017 visit produced $250bn in announced deals, most of which never materialized.
WHAT CHINA WANTS FROM NVIDIA
Chinese AI labs — DeepSeek, Moonshot, Alibaba's Qwen team — train on whatever Nvidia silicon they can legally import, plus stockpiled H100s smuggled through third countries. Huawei's Ascend chips work but lag a generation behind, and the CUDA software ecosystem has no Chinese equivalent.
THE SECURITY ARGUMENT VS THE REVENUE ARGUMENT
The Pentagon view: every advanced chip sold to China shortens the timeline to military AI parity. The Treasury and Commerce view: cutting Nvidia off from a $50bn market hands that market to Huawei and accelerates Chinese self-sufficiency. The trade-off has no equilibrium — each side's argument gets stronger as the other side's policy succeeds.
THE LEVERAGE QUESTION
Trump frames chip access as a bargaining chip — open Chinese markets to US firms or the export controls stay. But the controls were originally framed as national security, not trade. Trading them for market access concedes that the security rationale was negotiable all along.