[News] U.S. AI Chip Controls Are Coming— A Breakdown of the Tiered Restrictions

The U.S. government’s new tiered AI chip export restrictions—known as the AI Diffusion Rule—are set to officially take effect on May 15. According to Liberty Times, the regulation is expected to significantly impact major U.S. tech companies’ plans to invest in overseas data centers.

As highlighted in the report, the U.S.’s three-tiered control framework now extends to most countries and regions around the world. Each tier imposes different limits on AI compute deployment. Below is a breakdown of each tier, based on analysis from the American think tank Center for Strategic and International Studies (CSIS).

Tier 1 (T1): Trusted Allies with Conditions

Tier 1 consists mostly of long-standing U.S. allies—18 countries grouped into three main categories:

Five Eyes intelligence partners: Australia, Canada, New Zealand, and the United Kingdom

Close Western/NATO allies: Belgium, Denmark, Finland, France, Germany, Ireland, Italy, Netherlands, Norway, Spain, Sweden, and the United Kingdom

Major semiconductor players: Taiwan, the Netherlands, Japan, and South Korea

According to CSIS, Tier 1 countries will have unrestricted access to advanced GPUs. Companies that are either headquartered in, or ultimately owned by entities in, T1 nations are allowed to deploy unlimited AI computing power within other T1 countries.

Additionally, through a one-time Universal Validated End User (UVEU) authorization, they may also deploy GPUs in Tier 2 countries, though this is subject to specific limitations.

As the report highlights, Tier 1 status is not without limitations. U.S.-headquartered companies are required to keep at least 50% of their total AI computing power within the U.S., ensure that at least 75% stays within Tier 1 countries, and restrict deployment in any single Tier 2 country to no more than 7%.

Tier 2 (T2): Including Most Countries, Strict Limits

Tier 2 includes a wide range of countries such as India, Israel, Singapore, Switzerland, and Yemen. As the report points out, Saudi Arabia and the United Arab Emirates, despite emerging as major investors in AI, are still categorized as Tier 2.

As per CSIS, each Tier 2 country is allocated a total of 49,901 H100-equivalent GPUs through 2027. For smaller-scale needs, Tier 2 entities are allowed to acquire up to 1,699 H100-equivalents without a license, with prior notification required. However, CSIS notes that these caps may effectively shrink over time due to improvements in hardware performance.

The report highlights that Tier 2 countries are subject to rules that limit AI compute deployments to no more than 7% of a company’s global total. This restriction presents a significant challenge for U.S. tech giants such as Oracle, which have been heavily investing in overseas data centers. According to the report, these new rules are expected to substantially hinder Tier 2 countries’ ability to attract future AI-related investments.

Tier 3 (T3): Restricted and Sanctioned States

Tier 3 includes countries typically under U.S. arms embargoes, such as China, Iran, North Korea, Russia, Myanmar, Syria, and Venezuela. These countries are effectively barred from accessing advanced U.S. AI compute capabilities.

Tech Giants Lobby for Flexibility on AI Restrictions

The report, citing an analyst, emphasizes that the U.S. remains the world’s leading developer of AI technologies, with over USD 328 billion invested in the field over the past five years. Meanwhile, according to Bloomberg, major tech companies such as NVIDIA and Oracle are actively lobbying the Trump administration to ease the forthcoming restrictions, aiming to safeguard their global AI strategies.

Read more

• [News] Biden’s AI Chip Export Curbs in the Works: Key Points of Three-Tier System at a Glance
• [News] China’s AI Compute Dilemma: Why Advanced GPUs Are Sitting Unused in Idle Data Centers?