H100 GPU from NVIDIA
The Trump administration is signaling potential revisions to the Biden-era “AI Diffusion” export control framework. The shift could mean simpler, bilateral agreements and easing restrictions for key markets, following intense lobbying from U.S. tech giants and mounting economic concerns. Reuters was among the first to break the news.
President Trump has lent weight to these considerations, hinting that announcement may come soon.
This potential overhaul, which officials are reportedly considering before or shortly after the original framework’s May 15 effective date, aims to reshape how the U.S. balances national security with the economic imperatives of its leading semiconductor industry. In particular, the shift could benefit nations like the UAE and Saudi Arabia, and potentially India, by moving away from broad-stroke limitations.
Yet the potential easing for allies contrasts with the administration’s continued hard line on China. In April, Washington moved to indefinitely restrict sales of Nvidia’s H20 AI chip to China, as The New York Times and others reported. NVIDIA specifically designed the H20 to comply with earlier rules. The fallout from that decision led t an anticipated a $5.5 billion financial charge for the company.
The Biden administration had moved to limit the global sale of artificial intelligence (AI) chips in January. At that point, Ken Glueck, executive vice president at Oracle, called the regulation “the mother of all regulations.” Others in the tech and semiconductor industry reacted similarly to the proposal. An NVIDIA exec called the rules “misguided” and “sweeping overreach.”
The rules split countries into three tiers with differing access levels.
In Tier 1, which included allies such as Australia, Canada, Japan, South Korea, Taiwan, and most NATO countries, were exempt from any caps and could receive unlimited AI chips. Tier 2, however, which included some 120 countries, had quantitative limits (“caps”) on how many high-end AI chips they could import. It included significant partners such as India, Israel, Saudi Arabia, the UAE, Singapore, Vietnam, and Mexico. Through 2027, each Tier 2 country was capped at a total compute power equivalent to 790 million in “total processing performance” (TPP), roughly equal to 50,000 Nvidia H100 GPU. Tier 2 caps were structured to encourage compliance mechanisms: for example, verified “end users” in these countries (such as trusted cloud providers) could be granted higher quotas (up to approximately 320,000 GPUs over two years, according to Reuters. Additionally, top U.S. cloud firms operating globally could become “Universal Verified End Users” exempt from caps. Finally, Tier 3 comprised “Countries of Concern,” a short list of adversarial states, including China, Russia, Iran, and North Korea. Those were outright blocked from obtaining the covered AI chips.
