US Preps Sweeping AI Chip Export Controls
The U.S. government is reportedly drafting new rules that would require a license for nearly all global exports of advanced AI accelerators from companies like Nvidia and AMD. This move would expand controls beyond China to the entire world, potentially tying export permissions to investment and localization pledges, significantly tightening the leash on the global AI supply chain.
The proposed regulations introduce a tiered licensing system based on the scale of the chip deployment. Shipments of up to 1,000 GPUs, like Nvidia's GB300, would likely undergo a simplified review, but larger projects would face much stricter scrutiny and require pre-clearance from the Commerce Department. For massive AI clusters of 200,000 or more GPUs operated by a single entity, the host nation's government might need to negotiate directly with the U.S. and provide national security assurances. This represents a significant expansion from current controls that primarily target about 40 high-risk countries to a nearly global oversight model. This policy mirrors the approach recently used for AI semiconductor exports to Saudi Arabia and the UAE, where restrictions were eased after those nations pledged significant investments in the U.S. The draft rules suggest that for the largest deployments, "matching" investments in U.S. AI infrastructure could become a prerequisite for an export license. The Commerce Department, under the Trump administration, has confirmed it is formalizing this new approach, which it contrasts with the "burdensome, overreaching, and disastrous" AI Diffusion Rule from the previous Biden administration that was rescinded in May 2025. The stated goal is to position the U.S. as a central gatekeeper for the global AI industry, ensuring the American tech stack remains foundational to worldwide AI development.
