U.S. Rule Limits Cloud Access to Advanced Chips at 12 Million Compute Points
Key Facts
The U.S. Commerce Department’s Bureau of Industry and Security issued an interim final rule regulating remote access to advanced computing chips and servers, extending export controls to cover the use of computing power through data centers. The bureau published a notice announcing the rule on November 15, 2026, while The Associated Press reported that the Commerce Department issued it on Monday, November 16. The bureau said the rule was prepared at the direction of the Treasury Department and the National Security Council. It aims to align safeguards for remote access with existing restrictions on the physical shipment of sensitive U.S. technology. Its main threshold is more than 12,000,000 aggregate performance points when the service is provided to companies headquartered in Macau or a destination in Country Group D:5.
The rule targets computing arrangements that allow customers in covered destinations to use advanced capacity from a data center located outside those destinations. Covered customers include a company headquartered in Macau or a Country Group D:5 destination, as well as a company whose ultimate parent is headquartered there. The rule may also apply if a customer outside those destinations makes computing capacity available to a company in the group for model training or provisioning. Under the rule’s text, certain items subject to export controls may not be provided without a license when the service provider knows they will serve a data center that makes the specified capacity available. The review is therefore tied to the identity of the ultimate beneficiary and how the service is used, as well as the data center’s location.
The threshold is not limited to a single facility; the capacity of interconnected data centers can be aggregated to reach it. The text defines interconnection to include digitally connected infrastructure, not only physically connected facilities. This brings distributed operating arrangements within the scope of assessment when they serve the same customer. The bureau requires foreign data centers to determine whether people from Macau or Country Group D:5 own or control the customer, directly or indirectly. This includes attention to corporate restructuring or separation that could be used to circumvent the restrictions.
The bureau justifies expanding oversight on the grounds that access to this scale of computing could accelerate the training or operation of advanced models, raising national security and foreign policy risks. The text cites the possibility that these capabilities could be used to develop weapons of mass destruction or carry out sophisticated cyberattacks targeting critical infrastructure. The rule describes addressing this technological development as aligning safeguards for remotely available computing power with controls on shipments of sensitive technology. This reflects a shift in part of the oversight toward the service provided by chips, rather than only the movement of physical chips. A service provider’s knowledge of who will use the data center remains part of the licensing standard set out in the rule.
The rule also requires U.S. licenses for the most advanced chips sent to Singapore, Thailand, Malaysia, Indonesia and Brazil, according to Reuters’ report on the measure. This adds five destinations to the picture of controls linked to chip exports, while the provision on remote access concerns data centers and the customers benefiting from their capacity. Operators providing capacity exclusively for qualifying commercial recommendation systems may continue service for 90 days after the rule takes effect, after which they will need a license to continue. A 90-day transition period also gives existing data centers that provide significant capacity to companies headquartered in Macau or Country Group D:5 time to obtain the license needed to continue service. The text ties the continuation of some existing services to completing licensing requirements within that period.
The announcement came on a day when U.S. stocks fell in trading, amid pressure on technology shares and concerns about valuations in the artificial intelligence sector and the possibility that it was entering a bubble, according to The Associated Press. As of 10:15 a.m. Eastern Time on November 16, the S&P 500 was down 1.3%, the Nasdaq Composite had fallen 2.1%, and the Dow Jones was down 332 points, or 0.6%. Broadcom and AMD shares each fell 4.7%, while CoreWeave dropped 6.4% in the same report. The agency also linked the pressure to concerns about possible restrictions on building additional data centers after the U.S. midterm elections. Nvidia shares had closed at $162.85 on November 13, according to EL7 data.
The Associated Press identified Nvidia’s quarterly results, scheduled to be announced after the close of U.S. trading on Tuesday, as the next major event for the artificial intelligence sector. The company’s interim chief executive, Colette Kress, was expected to speak to investors. Reuters also reported that President Donald Trump had begun a visit to China and was scheduled to meet President Xi Jinping on Thursday in Shenzhen during the APEC summit. These dates put company results and trade developments at the forefront of attention during the same week. For existing data centers covered by the rule, continued service is tied to the specified period for obtaining licenses after it takes effect.