
AI Diffusion Rule
The AI Diffusion Rule is a US export regulation that governs which countries are allowed to purchase powerful AI chips and AI models. It aims to prevent technology needed to train powerful AI systems from falling into the hands of states that the US classifies as security risks.
The AI Diffusion Rule is a US government regulation that restricts the export of certain computer chips and AI technology abroad. It was issued in January 2025 under President Biden and took effect shortly thereafter. The rule divides all countries in the world into three groups: close allies with few restrictions, neutral states with stricter caps, and countries like China or Russia, to which such technology may generally not be sold at all. This makes it one of the most far-reaching measures with which the US is attempting to secure its leading role in artificial intelligence. The responsible agency is the Bureau of Industry and Security, or BIS for short, which falls under the US Department of Commerce.
Strategic significance for global AI competition
Powerful AI systems do not appear out of thin air. They require specialized chips — above all so-called GPUs, graphics processing units designed for parallel computation. Whoever has these chips can train AI. Whoever cannot obtain them can hardly do so.
This is exactly where the rule comes in. The US is the most important producer of these chips, above all through the company Nvidia. By controlling exports, Washington indirectly controls who is even capable of developing competitive AI systems. This is not a technical detail but a geopolitical decision: which countries are allowed to ride along at the forefront of AI development — and which are not?
A common misconception is to read the rule as a purely economic sanction. Officially, it is about national security: military applications, surveillance systems, and autonomous weapons could benefit from the same AI models that today write texts or generate images. The dividing line between civilian and military AI is fluid — which makes the control especially politically sensitive.
The rule’s three-tier system
The rule sorts countries into three categories, often referred to in technical documents as Tier 1, Tier 2, and Tier 3. Tier 1 includes close allies such as Germany, Japan, or the United Kingdom. They may obtain chips and models largely without quantity restrictions. Tier 2 comprises neutral or less closely allied states, such as India or Mexico. They may buy, but only up to certain caps. Tier 3 — which includes China, Russia, Iran, and North Korea — is largely locked out.
The caps for Tier 2 countries are measured in an abstract unit called Total Processing Performance, or TPP for short. This unit describes how much total computing power a shipment contains. Anyone wanting to purchase a large quantity of chips needs a special license above a certain threshold. This is meant to prevent countries from being used as a detour to resell the technology afterward.
In addition to chips, the rule also covers model weights. These are the stored results of an AI training process — put simply, the actual “intelligence” of a model. Their export is also subject to a licensing requirement above a certain performance threshold. This is new and shows that the government now treats AI software as a security good just as much as physical hardware.
Where the AI Diffusion Rule appears in the news
The rule has been a constant topic in the tech and financial press since its announcement. Nvidia shareholders watch it closely, because a significant portion of Nvidia’s revenue comes from regions that fall under Tier 2 or are disputed. Any tightening of the rule can immediately move the stock price.
Countries seeking to build large data centers are also directly affected. Saudi Arabia, the United Arab Emirates, and other Gulf states have ambitious AI plans — and fall under Tier 2. They must individually prove that their infrastructure is secure enough to receive larger quantities of chips. This leads to diplomatic negotiations that go far beyond normal trade agreements.
The Trump administration, which took office in early 2025, initially announced it would review the rule. This shows how politically charged the topic is: what counts as a security measure under one administration can be assessed as a trade barrier under another. Anyone reading tech news will therefore still encounter the term often — whenever the competition between the US and China in AI is at stake.