• The Commerce Department will require licenses for remote access to US AI chips when computing power exceeds 12 million TPP, roughly 750 Nvidia (NASDAQ:NVDA) H200 chips.
  • New licensing requirements also target chip shipments to Indonesia, Malaysia, Singapore, Thailand, and Brazil to curb diversion risks.
  • The interim final rule, set for Federal Register publication on November 19, formally rescinds the Biden administration's AI Diffusion Rule and adds supply-chain screening obligations.

The US Commerce Department is establishing new rules to prevent certain foreign actors from leveraging advanced computing power based on US technology to threaten national security and foreign policy interests, according to a Federal Register notice. The interim final rule, announced by the Bureau of Industry and Security, targets a gap in existing controls: companies that cannot import the most advanced US chips can still rent computing power from those chips in overseas data centers.

As of the morning of November 18, the rule was on public inspection and scheduled to take effect upon Federal Register publication on November 19, with a separate 90-day transition for certain existing arrangements. It formally rescinds the Biden administration's broader AI Diffusion Rule.

Remote Access and Shipment Controls

The rule requires a license for covered exports, reexports, or in-country transfers when the supplier knows the chips or systems are destined for a data center that provides, or will provide, more than 12 million total processing performance (TPP) to or for covered customers. TPP is the rule's measure of chip computing power; the threshold is roughly equivalent to 750 Nvidia H200 chips. The rule reaches linked data centers, intermediaries, and customers whose ultimate parent is headquartered in Macau or a covered US arms-embargoed jurisdiction, including China.

Separate licensing requirements apply to covered chips and systems going to or transferred within Indonesia, Malaysia, Singapore, Thailand, and Brazil. BIS identified those destinations as presenting a heightened risk of diversion. This is a licensing requirement, not a blanket embargo; the rule also makes a limited exception for computing used exclusively for commercial recommendation systems, subject to conditions.

The rule strengthens customer and ownership checks and adds licensing requirements for certain transactions by chip foundries and outsourced assembly and testing providers. The practical change is that chip suppliers and cloud operators need to pay closer attention to who ultimately uses the computing capacity, not just who buys or houses the hardware.

Nvidia in Focus

The rule applies across the industry, but Nvidia is the most prominent company exposed. The California-based semiconductor designer supplies graphics processors and data-center AI systems, including the H200, Blackwell, and Vera Rubin platforms. In results released November 17, it reported fiscal third-quarter revenue of $115.8 billion, up 103% year over year, and diluted earnings of $2.68 per share. Both beat expectations. Its fourth-quarter revenue outlook was about $130 billion, plus or minus 3%, with no China data-center chip sales assumed. The reported quarter ended before both the leadership change and the new rule, so it is not evidence of the rule's financial impact.

Co-founder Jensen Huang resigned as CEO, president, and director on November 12. CFO Colette Kress became interim CEO while retaining her finance role; Harvey C. Jones became board chair. The board is considering internal and external candidates for a permanent successor and said it does not intend to reinstate Huang. Nvidia declined to comment on the new rule, Reuters reported.

Political and Economic Context

The rule builds on US restrictions on advanced chip exports to China introduced in October 2022. As access to physical chips tightened, Southeast Asian data centers increasingly rented computing capacity to Chinese customers. The Biden administration's January 2025 AI Diffusion Rule took a broader approach, using country tiers and controls on certain model weights. The Trump administration announced in May 2025 that it would abandon that framework; the new rule formally removes its central provisions and instead focuses on specified hardware, end users, and destinations.

Washington says overseas computing access can support advanced AI development for military and intelligence uses. Beijing's commerce ministry rejected the restrictions on November 17 and threatened countermeasures if Washington proceeded, while calling for consultation. The announcement came as President Donald Trump was headed to China for talks with President Xi Jinping. In a separate policy, BIS has allowed case-by-case review of direct exports of Nvidia H200 and AMD (NASDAQ:AMD) MI325X chips to approved Chinese customers under specified security conditions. The two policies distinguish licensed direct sales of certain chips from remote access to large pools of controlled computing power.

What to Watch

Analysts surveyed by Reuters expect the first effects to show up in chip deployment, licensing delays, and customer verification—not necessarily in immediate construction cancellations. Projects serving established US cloud providers may be better positioned than facilities that depend on Chinese AI tenants or brokers. If delays persist, exposed operators could see lower utilization, pricing pressure, or postponed expansion.

Chinese AI developers could face higher computing costs and slower model-development work in the short term. Longer term, analysts see an incentive to invest in domestic chips and computing infrastructure, potentially deepening the separation between US and Chinese AI supply chains. That outcome depends in part on how consistently the rules are enforced.

The exception for qualifying recommendation systems draws a line between some everyday retail or social-media workloads and the large-scale AI workloads that concern BIS. The dispute is over how to enforce that line: US officials emphasize security risks, while Beijing argues the controls threaten semiconductor supply chains, and operators face the practical burden of verifying customers and workloads.

The clearest near-term indicators are how BIS handles licenses during the transition, whether Beijing follows through on its warning, and whether operators in the five named countries delay bringing new computing capacity online. Those developments will show whether the immediate effect is mainly higher compliance costs or a meaningful shift in where AI computing capacity is built and who can rent it.