Technology

How AI Chip Export Controls Work—and Why They Matter

The U.S. uses a tiered export control system to restrict which countries can buy advanced AI chips, reshaping global technology competition and sparking a cat-and-mouse game with smugglers.

R
Redakcia
4 min read
Share
How AI Chip Export Controls Work—and Why They Matter

Why Chips Became a Weapon

Advanced artificial intelligence runs on specialized processors—graphics processing units (GPUs) and purpose-built accelerators that can perform trillions of calculations per second. Whoever controls access to these chips shapes who can train the most powerful AI models. That reality turned semiconductors into a front line of geopolitical competition.

Since October 2022, the United States has built an increasingly complex system of export controls designed to keep cutting-edge AI chips out of adversaries' hands, primarily China's. Administered by the Bureau of Industry and Security (BIS) within the Department of Commerce, these rules determine which chips can be sold, to whom, and under what conditions.

The Three-Tier Framework

The centerpiece of U.S. policy is the AI Diffusion Framework, which sorts every country into one of three tiers based on strategic alignment and national-security risk.

  • Tier 1 — The U.S. and 18 close allies (Five Eyes members, key NATO partners, semiconductor-ecosystem nations like the Netherlands and Taiwan). These countries face no import limits on advanced AI chips.
  • Tier 2 — Most other nations. They can import limited quantities—roughly 50,000 H100-equivalent GPUs per country through 2027—unless they join a special data-center authorization program or obtain individual licenses from BIS.
  • Tier 3 — Arms-embargoed countries including China, Russia, Iran, and North Korea. Exports of advanced AI chips are effectively banned.

The framework also introduced, for the first time, controls on AI model weights—the trained parameters of large frontier models—requiring licenses to export closed-source weights above a computational training threshold of 1026 operations.

How BIS Decides What's Restricted

BIS classifies chips using performance thresholds measured in total processing power. When Nvidia designed slightly slower chips specifically for the Chinese market after the initial 2022 rules, BIS closed that loophole in October 2023, noting the workaround designs still offered "nearly comparable AI model training capability." Subsequent updates in 2024 expanded controls to cover high-bandwidth memory (HBM), advanced chip packaging equipment, and DRAM components critical to AI systems.

Companies that want to export restricted items must apply for a license. BIS reviews each application on a case-by-case basis, weighing national-security risk against commercial impact. The Entity List—a roster of foreign organizations deemed threats—triggers a presumption of denial. BIS has added more than 200 Chinese entities to the list since 2022, including AI labs, chipmakers, and military-linked research institutes.

The Smuggling Problem

Export controls are only as good as their enforcement, and a growing body of evidence shows significant leakage. In March 2026, federal prosecutors charged a Super Micro Computer executive with diverting Nvidia-equipped servers to China through shell companies in Taiwan and Southeast Asia, using falsified documents to pass compliance audits. In a separate case, three individuals were arrested for ordering 750 servers worth approximately $170 million under fraudulent end-user certifications.

Circumvention has also taken creative forms. Rather than smuggling hardware, some Chinese companies have begun exporting data to locations where advanced chips remain accessible—effectively bringing the training workload to the chips instead of the other way around.

BIS faces a resource mismatch: its enforcement budget has not kept pace with the vastly expanded scope of its mandate since 2022, according to a Bloomsbury Intelligence and Security Institute report.

The Chip Security Act

Congress has responded with new legislative tools. The Chip Security Act, advanced for a full House vote in March 2026, would require tracking technology to be embedded directly on exported chips, allowing BIS to verify a chip's location and ownership after sale. While not foolproof—chips could theoretically be taken offline or their signals spoofed—proponents argue even partial visibility would be a dramatic improvement over the current paper-based licensing system.

Why It All Matters

AI chip export controls represent a historic bet: that controlling hardware can slow a rival's AI progress without crippling legitimate global commerce. Critics warn the restrictions push allies toward building independent supply chains and risk fragmenting the global semiconductor ecosystem. Supporters counter that without controls, adversaries would rapidly close the capability gap in military AI, surveillance, and cyber operations.

As AI grows more powerful—and the chips that drive it grow more strategically valuable—the architecture of these controls will shape not just trade policy, but the global balance of technological power for decades to come.

Stay updated!

Follow us on Facebook for the latest news and articles.

Follow us on Facebook

Related articles