The US government is moving to replace the current country-driven framework for advanced semiconductor exports, particularly those used in Artificial Intelligence, with a bilateral licensing system. Sources familiar with the deliberations report that the new arrangement would require government-to-government agreements, mandating bilateral approval from both the US and recipient nations before any advanced chip sale can proceed. This significant policy shift is poised to tighten oversight of US-developed Artificial Intelligence chip technologies and further restrict access for countries of security concern, such as China, Russia, Iran, and North Korea.
Under the current framework, in effect since January 2025, there are three tiers of export controls: unrestricted sales to 17 US allies and Taiwan, regulated volume caps for around 120 countries, and a complete export bar on the highest-risk nations. Presently, orders involving fewer than 1,700 NVIDIA H100 equivalent chips require only notification, rather than full government licensing. However, to address persistent fears over circumvention and loopholes, officials are also contemplating lowering this threshold to about 500 units, vastly expanding the net of regulated exports.
The potential overhaul arrives amid mounting industry and legislative scrutiny. Former Commerce Secretary Wilbur Ross, now serving as an informal adviser, confirmed that bilateral licensing agreements are actively being considered. High-profile criticism has surfaced from industry leaders like Oracle Executive VP Ken Glueck, as well as a group of seven Republican senators, all of whom have lobbied Commerce Secretary Howard Lutnick to retract the existing export framework. With a critical compliance deadline of May 15 approaching, the administration is racing to reconcile trade interests with national security objectives. An official announcement detailing the new policy is anticipated before the close of the month.
