Guide · Markets
Export Controls Explained: How US Chip Curbs Work
Export controls are rules that require a government licence to sell certain goods abroad. The US uses them to keep advanced AI chips away from China.
On Monday, November 16, 2026, the US Commerce Department issued a rule requiring a licence to ship advanced AI chips to any data centre in the world that rents large amounts of computing power to companies headquartered in China or other restricted countries. Nvidia fell 4.12% that day and Asian chip stocks slid the next morning. To understand why a few pages in the Federal Register can move trillion-dollar companies, it helps to know how export controls work.
Export controls are laws that require a government licence before certain goods, software or technology leave the country, or reach certain buyers. In the United States, most dual-use items, the ones with both civilian and military uses, fall under the Export Administration Regulations, run by the Bureau of Industry and Security inside the Commerce Department. Every controlled item has a classification number on the Commerce Control List; advanced AI chips sit under 3A090 and the computers that contain them under 4A090. Whether a licence is needed depends on three things: what the item is, where it is going and who will use it.
The modern chip controls date from October 7, 2022, when BIS cut China off from Nvidia's A100 and H100 and from the equipment needed to make leading-edge chips. Nvidia responded with slower versions, the A800 and H800, designed to fall just under the limits. A year later, on October 17, 2023, BIS rewrote the test around total processing performance, or TPP, a single number that combines a chip's speed with the precision of its calculations, and caught those variants too. Nvidia then built the H20. On April 15, 2025, the company said the US now required a licence to sell the H20 to China and that it expected charges of about $5.5 billion as a result. Each round followed the same pattern: a rule, a redesigned product, then a tighter rule.
Three tools do most of the work. The first is the licence requirement itself, often with a "presumption of denial" for the most sensitive destinations, which means applications are expected to be refused. The second is the Entity List, a register of named companies, universities and individuals that need a licence for almost anything, used against firms suspected of diverting goods. The third is the foreign direct product rule, which extends US controls to items made abroad with American technology or tools, the reason a chip made in Taiwan can still fall under US law. The November 2026 rule added a fourth angle: controlling where chips are installed and who rents them, not only who buys them.
For investors, the effects show up in three places. Revenue: Nvidia stopped reporting meaningful data-centre sales to China after successive rules, and every new control raises the question of how much more demand is fenced off. Inventory and charges: when a product is suddenly blocked, unsold stock and purchase commitments can turn into write-downs, as the 2025 H20 charges showed. And competition: controls protect a technological lead but also push Chinese buyers towards domestic chips. Rules usually appear without warning, often as interim final rules that take effect before the comment period ends, which is why chip stocks can gap on a Monday morning.
To follow the story, watch the Federal Register and BIS press releases for new rules and Entity List additions, company filings for licence disclosures and charges, and the countries named as diversion risks, which in November 2026 included Malaysia, Singapore, Indonesia, Thailand and Brazil. A rule that is easy to work around tends to fade from prices quickly; one that targets the workaround itself, as the November 16 rule tries to, deserves more attention.