The Trump administration is weighing a new approach to AI chip exports, and early signals suggest it could be much tougher than the semiconductor industry expected. If the proposal moves forward, it would reshape how cutting-edge AI hardware from companies like NVIDIA and AMD is allowed to ship overseas, with new rules that focus heavily on the amount of computing power leaving the United States.
AI export controls have been a recurring flashpoint ever since major chipmakers delivered huge leaps in performance that made their accelerators essential for training and running advanced AI models. The issue gained major traction under the previous administration, which introduced a framework often described as an “AI diffusion” policy. That earlier approach focused on grouping countries into tiers, with different restrictions depending on the destination. It drew criticism across the AI supply chain, including objections from NVIDIA, as many saw it as a potential drag on America’s competitiveness.
Now, reporting indicates the Trump administration may be considering regulations that go further by tightening export licensing requirements across the board. The key change: instead of targeting a limited set of countries, the U.S. could require export licenses for AI hardware shipments to any country worldwide, potentially including close allies. As described, there would be no broad exemptions, meaning manufacturers could face a more universal approval process before shipping advanced AI systems abroad.
A major factor in how strict the review process would be appears to be compute power. In other words, regulators may evaluate exports by measuring the scale of AI capability being shipped. Under the reported concept, smaller shipments—up to the equivalent of 1,000 NVIDIA GB300 racks—could go through a relatively straightforward review. Larger deployments, however, could trigger a much more demanding process.
For big, data-center-scale orders, the proposal would reportedly require involvement from the destination country’s government, with negotiations led at the state level rather than handled purely between companies. The stated expectations could include security-related commitments and investments that align with U.S. interests in AI—effectively tying access to American AI hardware to policy assurances and economic participation.
Supporters may argue this creates a clearer framework for protecting national security and controlling the spread of highly capable AI infrastructure. Critics, however, are likely to point to the same concern raised in past debates: that a licensing-heavy model could slow down global AI buildouts through delays, paperwork, and uncertainty—especially when approvals are needed for multiple markets and projects at once.
So far, there’s no confirmed timeline for when such rules would be finalized or implemented. With the earlier AI diffusion rule having been rescinded in May 2025, it’s possible a new set of restrictions could arrive on a similar policy cadence, but the details—and the real-world impact on AI chip supply chains—remain to be seen.
