US Commerce Department codifies H200-to-China export rule

The US Bureau of Industry and Security published a rule putting into regulatory form the policy Donald Trump had announced the previous December: allowing Nvidia to sell H200 chips, and AMD its comparable MI325X, to approved customers in China under a licensing regime, moving the chips out of the near-blanket presumption of denial that had applied to China’s most advanced AI hardware. Trump followed with a formal proclamation the next day.

The rule capped exports of each chip at 50% of the exporter’s cumulative US sales of that same product, a mechanism intended to ensure American buyers were served first. Outside analysts calculated that threshold could allow China to acquire on the order of 850,000 H200s, or nearly 900,000 chips once AMD’s equivalent hardware was included — far short of the unrestricted access Nvidia had sought, but a marked loosening from the export ban that had been in place. Sales also carried a 25% fee on their dollar value, routed through mandatory third-party testing at US facilities before re-export, and exporters had to certify that shipments would not reduce the chip supply available to US customers and would not reach parties on restricted lists.

Under Secretary of Commerce Jeffrey Kessler said export controls should “evolve with changes in technology, while protecting national security.” Critics, including members of Congress who had pushed competing legislation to block the sales outright, argued the security conditions — chiefly the know-your-customer screening and end-use certifications — would be difficult to enforce against shell companies and China’s civil-military integration, leaving the policy’s practical effect on China’s frontier AI compute access uncertain even as the paperwork took effect.