For three years US policy on AI chips for China had one direction: tighter. In January 2026 it turned. Whether that changed much in practice is another matter.

The new rule

On January 13, 2026, the Commerce Department’s Bureau of Industry and Security announced that licence applications for chips like Nvidia’s H200 and AMD’s MI325X would move from a “presumption of denial” to case-by-case review for China and Macau. The rule, published on January 15, followed President Trump’s announcement in December 2025 that H200 sales would be allowed.

The conditions, from the rule text:

  • Performance limits: chips must fall below a total processing performance of 21,000 and total DRAM bandwidth of 6,500 GB/s.
  • Volume cap: shipments to China and Macau are limited to 50% of the volume shipped to US customers.
  • Testing and screening: independent third-party testing in the US, and customer screening to keep chips away from military users.
  • A cut for Washington: the US takes 25% of the sales, which law firms describe as implemented through a tariff proclamation.

Beijing’s answer

China did not rush to buy. Reuters reported in January that Chinese customs had been told to block H200 imports and that companies were advised not to buy them unless necessary, according to CNBC. China has not confirmed that instruction.

The effect showed up in Nvidia’s numbers. On its February earnings call, CFO Colette Kress said small amounts had been approved but there was no revenue yet, and that Nvidia did not know whether China would allow any imports. By July a Commerce official told Congress that H200 shipments had been trivial against roughly $10 billion in licences. Nvidia has excluded China data-centre sales from its forecasts all year.

The wider trade picture

Chips were one piece of a bigger negotiation:

  • Congress pushed back. The House Foreign Affairs Committee advanced the AI Overwatch Act, which would require congressional review of chip export licences.
  • The leaders met. Trump and Xi met in Beijing on May 14. US trade officials said export controls didn’t come up, though Reuters reported H200 sales to several Chinese firms were cleared around then.
  • Rare earths stay on a short leash. On November 9 China extended its pause on expanded rare-earth export controls to January 10, 2027, continuing the truce agreed in Busan in October 2025. China produced about 91% of refined rare earths in 2024, according to the IEA, as cited by The National.

Why it matters

  • US policy changed from denial to licensed, taxed exports. That is a new model for technology controls, and one Congress may yet limit.
  • China is choosing self-reliance. By discouraging H200 purchases, Beijing is pushing its AI companies toward domestic accelerators, which shapes the hardware Chinese open-weight models are trained on.
  • Hardware supply chains run on two-month extensions. Rare-earth access now depends on rolling truces, a real risk for anyone building physical products.

Updated November 9, 2026 with China’s rare-earth extension.

Sources

  1. BIS: Commerce revises license review policy for semiconductors exported to China (Jan 13, 2026)
  2. Federal Register: rule 2026-00789, public inspection copy (Jan 15, 2026)
  3. CNBC: Trump, Nvidia AI chip exports to China and a congressional bill (Jan 22, 2026)
  4. CNBC: Nvidia China chip sales and export controls (Feb 26, 2026)
  5. The National: China extends pause on rare-earth export curbs for two months (Nov 9, 2026)

We link every claim to the reporting or official statement it comes from. Where a detail rests on one outlet’s reporting, we say so. Spotted an error? Reply to any newsletter and we’ll correct it.