The Biden administration has announced a major export control plan meant to keep advanced AI chips and the most powerful AI models away from adversaries such as China. The measure, called the AI Diffusion rule, would divide global access to US AI technology into trusted routes, licensed routes, and already restricted destinations.
The proposal reaches beyond hardware. For the first time, it also seeks to control access to advanced model weights, the technical core of closed AI models made by US firms.
What the AI Diffusion rule changes
The rule creates a new framework for deciding who can access America’s most advanced AI silicon and algorithms. It will be enforced by the Commerce Department’s Bureau of Industry and Security.
At the center of the plan is a distinction between countries that are allowed relatively broad access and countries where companies will need special permission for larger deployments or more sensitive technology. US Commerce Secretary Gina Raimondo framed the rule as a way to preserve America’s lead in AI development and AI chip design.
The trusted nations named in the rule are the UK, Canada, Australia, Japan, France, Germany, Belgium, Denmark, Finland, Ireland, Italy, the Netherlands, New Zealand, Norway, Republic of Korea, Spain, Sweden, and Taiwan.
Companies in other nations not subject to arms controls would still have access to up to 1,700 of the latest AI chips without special permission. Beyond that, they would need to apply for a special license.
Where licenses would be needed
The licensing requirement would apply when companies seek more chips, want to build very large scale data centers using US technology, or need access to the most powerful closed model weights made by US firms. To receive a license, companies would have to show adequate physical and cybersecurity.
That makes the rule about more than a shipment of hardware. It treats advanced AI capability as something created through a combination of chips, data centers, security controls, and model access.
The rule also draws clear lines around what it does not cover. Supply chain activities, including the design, manufacturing, and storage of chips, are exempt. Open source AI models such as Meta’s Llama are also not restricted under the administration’s description of the rule.
Arms-embargoed nations such as China, Iran, and North Korea are already barred from obtaining advanced chips. The new element is the move to restrict their access to advanced models as well.
Why China is the focus
The US already limits exports of advanced AI chips to China. According to the source article, companies there have still been able to build cutting algorithms using computer clusters located in other nations.
The AI Diffusion rule is designed to close that path. Under the rule, China would not be able to build so-called frontier AI models in other nations affected by the restrictions.
The Biden administration’s concern is that the same technologies powering commercial AI can also support national security threats. Raimondo described semiconductors that power AI and model weights as dual use technology. The administration says adversaries could use them for nuclear simulations, bio weapons development, and military advances.
Jake Sullivan, the Biden administration’s national security adviser, said the US must prepare for rapid increases in AI capabilities in the coming years, with possible effects on both the economy and national security.
A senior administration official who asked not to be named said the administration believes China’s AI development is between six and 18 months behind that of the US. The official said, "Time is really of the essence," and added, "We believe we're in a critical window right now."
Industry pushback and political timing
The rule is expected to be controversial because it could slow international AI sales at an important moment for the industry. It was announced just a week before Donald Trump’s inauguration.
The rule includes a 120-day consultation period. That means Trump‘s administration will be expected to hear input, potentially revise the rule, and then enforce it.
Nvidia, described in the source article as the world’s leading manufacturer of AI chips, criticized the rule in a blog post, calling it "unprecedented and misguided." The company argued that the measure would not improve US security and would weaken America’s global competitiveness.
The disagreement captures the central tension in the policy. The Biden administration is trying to prevent adversaries from reaching advanced AI capability through indirect channels. Nvidia warns that broad restrictions could harm the same US-led AI ecosystem the rule is meant to protect.
The broader export control push
The AI Diffusion rule is part of an escalating set of export controls aimed at limiting China’s access to cutting-edge AI. In December, the Biden administration introduced new restrictions on the export of chip-manufacturing equipment, certain types of computer memory, and chip design software to China.
Those measures were intended to weaken China’s ability to make advanced chips. The new rule shifts attention to the wider system around AI: chips, data centers, model weights, and the countries through which access can flow.
If enforced, the rule would make global AI expansion more dependent on licensing decisions and security requirements. It would also mark a significant step in treating powerful AI models themselves as controlled technology, not only the chips used to train and run them.