The U.S. wants to curb the sale of AI chips used in data centers on both a country and company basis, with the goal of concentrating AI development in friendly nations and getting businesses around the world to align with American standards, according to people familiar with the matter.

The result would be an expansion of semiconductor trade restrictions to most of the world—an attempt to control the spread of AI technology at a time of soaring demand. The regulations, which could be issued as soon as Friday, would create three tiers of chip curbs, said the people, who asked not to be identified because the discussions are private.

At the top level, a small number of U.S. allies would maintain essentially unmitigated access to American chips. A group of adversaries, meanwhile, would be effectively blocked from importing the semiconductors. And the vast majority of the world would face limits on the total computing power that can go to one country. 

Companies headquartered in nations in that last group would be able to bypass their national limits—and get their own, significantly higher caps—by agreeing to a set of U.S. government security requirements and human rights standards, according to the people. That type of designation—called a validated end user, or VEU—aims to create a set of trusted entities that develop and deploy AI in secure environments around the world.

Shares of Nvidia, the leading maker of AI chips, dipped more than 1% in late trading after Bloomberg reported on the plan. They had been up 4.3% this year through the close, following stratospheric gains in 2023 and 2024 that turned the company into the world’s most valuable chipmaker.

Advanced Micro Devices Inc., Nvidia’s biggest challenger in AI processors, dropped less than 1% in extended trading.

Nvidia objected to the proposal in a statement. “A last-minute rule restricting exports to most of the world would be a major shift in policy that would not reduce the risk of misuse but would threaten economic growth and U.S. leadership,” Nvidia said. “The worldwide interest in accelerated computing for everyday applications is a tremendous opportunity for the U.S. to cultivate, promoting the economy and adding U.S. jobs.”

A representative of the White House’s National Security Council declined to comment. The Commerce Department’s Bureau of Industry and Security, which is in charge of chip export controls, didn’t immediately respond to a request for comment.

The measures build on years of curbs that already limit the ability of American chipmakers like Nvidia and AMD to sell advanced processors in China and Russia. The U.S. also has sought to prevent adversary nations from accessing cutting-edge technology through intermediaries in places like the Middle East and Southeast Asia. The latest draft rules are part of that global effort.

The rules follow months of debate over how quickly and broadly to deploy U.S. chips to global data centers. Because American chips far outperform Chinese ones at AI tasks, companies and entire countries have indicated that they’re willing to jump through hoops to gain access to U.S. technology. That gives the U.S. a unique role as a gatekeeper—and a potentially significant amount of leverage to shape global AI development. 

The U.S. has “a serious once-in-a-generation moment to leverage U.S. AI technology,” the top Democrat and Republican on the House China Select Committee wrote last week in a letter to Commerce Secretary Gina Raimondo. “Demand for U.S. AI technology is an opportunity to pry both companies and countries out of Beijing’s orbit.” 

The first tier established in the new rules includes the U.S. and 18 allies, such as Germany, the Netherlands, Japan, South Korea and Taiwan, according to people familiar with the matter. 

Companies can freely deploy computing power in those places, and firms headquartered there can apply for blanket U.S. government permission to ship chips to data centers in most other parts of the world. That’s provided that no more than a quarter of their total computing power is located outside of Tier 1 countries, and no more than 7% in any one Tier 2 country. Companies would also have to abide by U.S. government security requirements.

Additionally, U.S.-headquartered companies that apply for that type of permission—a so-called universal VEU designation—have to keep at least half of their total computing power on American soil, people familiar with the matter said. The broader goal of these regulations is ensuring that the U.S. and allied countries always have more computing power than the rest of the world.

Restrictive tiers

The vast majority of countries fall into the second tier of restrictions, which establishes maximum levels of computing power that can go to any one nation—equivalent to about 50,000 graphic processing units, or GPUs, from 2025 to 2027, the people said. But individual companies can access significantly higher limits—that grow over time—if they apply for VEU status in each country where they wish to build data centers. 

Getting that approval requires a demonstrated track record of meeting U.S. government security and human rights standards, or at least a credible plan for doing so. Security requirements span physical, cyber and personnel concerns. If companies obtain national VEU status, their chip imports won’t count against the maximum totals for that country—a measure to encourage firms to work with the U.S. government and adopt American AI standards.

The third, most restrictive tier affects China, Macau and all countries for which the U.S. maintains an arms embargo—about two dozen nations in total, according to the people. Shipments to data centers in those places are broadly prohibited. 

Like Nvidia, the Semiconductor Industry Association trade group opposes the move.

“A policy change of this scope and significance should not be rushed out the door during a period of presidential transition and without meaningful input from industry,” the association said in a statement. “Too much is at stake here to circumvent a deliberative process. Our country needs to get this right so we can compete and win globally.”

Model weights

In addition to the semiconductor controls, the new rules also limit the export of closed AI model weights, which are the numerical parameters that software uses to process data and make predictions or decisions. 

Companies would be prohibited from hosting powerful closed model weights in Tier 3 countries, like China and Russia, and would have to abide by security standards to host those weights in Tier 2 countries. That means the controls on model weights don’t apply to companies that obtain universal VEU status, one of the people said. 

Open weight models—which allow the public to access underlying code—aren’t affected by the rules, nor are closed models that are less powerful than an already-available open model. But if an AI company wants to fine-tune a general-purpose open weight model for a specific purpose, and that process uses a significant amount of computing power, they would need to apply for a U.S. government license to do so in a Tier 2 country.