China condemns US remote AI-compute controls and vows to protect its interests

CASRAI

Source published: Site published:

Original article · فارسی

Chinese Foreign Ministry spokesperson Guo Jiakun condemned new US rules on remote access to AI computing on 17 November 2026. He also promised that China would take action to protect its interests. His response challenges restrictions announced one day earlier on access to advanced chips from outside the United States. This has turned cross-border technology cooperation and the computing systems used by international research partnerships into a diplomatic dispute.

Beijing objects to limits on technology cooperation

At the ministry’s regular press conference, Guo accused the United States of using national security as a reason to expand its control. He said Washington was misusing export rules to interfere with China’s normal trade and technology cooperation with other countries. The official transcript from 17 November records his answer to a question about the Commerce Department’s temporary final rule, called Addressing Remote Access to Advanced Semiconductors.

Guo said the rule was aimed at China’s high-tech industry. He called it economic pressure and an attempt by the United States to exercise power beyond its borders. He argued that it went against agreements between the two countries. He also said it harmed the legitimate interests of Chinese companies and disrupted global supply chains. These are the Chinese government's stated objections to the rule.

He promised that China would take the steps needed to protect its legitimate rights and interests. This is Beijing’s response to the announced rule, which follows the US action described in CASRAI’s earlier report on BIS rules for remote access.

Independent reports show the different sides of the dispute

SBS reporter Kim Young-a’s report from 17 November describes Guo’s comments at the briefing and his objection to interference with trade and scientific and technological cooperation.

Reuters reporter Karen Freifeld, in a report published by The Jakarta Post, also records Guo’s promise. The report also describes a different reaction from the United States Congress. John Moolenaar, the Republican chair of the House Select Committee on China, welcomed the rule. He said it would close gaps in export controls.

The Bureau of Industry and Security's announcement dated 16 November says the policy is meant to stop restricted parties from getting access to sensitive US technology through data centers in other countries. BIS says remote cloud computing has made this possible even when the computer chips are physically located outside countries covered by US restrictions.

The rule covers computing access and where hardware is sent

The temporary final rule, RIN 0694-AK50, describes two separate controls that matter to research computing supply chains. New section 744.24 covers the export, reexport and transfer of certain advanced-computing products. These controls apply when a person knows that a data center will provide more than 12 million in total combined processing performance, or TPP, to or for companies covered by the rule.

The company test looks at whether a company is headquartered in Macau or a Country Group D:5 destination, or whether its ultimate parent company is headquartered there. Computing capacity can come from one data center or from several connected centers, including facilities that work together through digital connections. The rule's explanation also covers indirect arrangements. These include cases where a middle company trains AI models for a covered company, takes part in training for it, or provides models to it. Applications for licenses are reviewed one by one.

A separate part of the rule adds licensing requirements for certain advanced-computing products sent to Indonesia, Malaysia, Singapore, Thailand and Brazil. An exception may apply to recipients headquartered in the United States, or whose ultimate parent company is headquartered there, if they meet the stated reporting requirements. This exception has different conditions from those that apply to remote-access uses.

Section 744.24 also provides an exception, subject to reporting requirements, for computing used only for certain commercial recommendation systems. The definition does not include systems designed or used to create new content, even if they also make recommendations. This difference matters when deciding whether a service is a recommendation system or is being used for research on generative AI models.

Singapore is discussing how the rule will work

Singapore's separate response shows some of the questions other countries face when putting the rule into practice. In a joint statement from 16 November, Singapore's digital and trade ministries said they were speaking with affected companies and US authorities about the rule's scope and how it would be applied. They were also exploring ways for Singapore companies to have trusted access to advanced semiconductors.

The ministries said Singapore would protect legitimate companies' access to technology. However, they also rejected using a connection to Singapore to get around the laws of other countries. CASRAI's report on Singapore's response describes these discussions. Guo's later statement adds China's formal objection to the US rule.

What research purchasing and compliance teams can check

The following points are editorial analysis of the rule. For universities buying access to powerful computing chips, the practical checks include the type of hardware, the service provider, the provider's ultimate parent company, who will benefit from the service and how it will be used. A supplier's location alone does not answer the questions about company headquarters and intended use in section 744.24.

Purchasing records can therefore show the difference between buying equipment and buying a remote-computing service. For equipment, staff can identify who receives it, who will use it and whether the supplier plans to rely on an exception based on reporting requirements. For computing services, they can ask how computing capacity is shared across connected data centers and which companies benefit from the model-training work. These details help with an AI export-control assessment. They do not mean that every academic cloud account is automatically banned.

For an institution that runs computing systems or gives partners access to them, the notice and approval rules in section 744.24(e) mean that changes in how a service is used may matter throughout a project. A research security office can connect contract checks with access records when the people or companies benefiting from a project, or the way it is used, change. The rules based on a company's location and ownership should be kept separate from the nationality of an individual researcher.

Publication sets the timetable for putting the rule into effect

The copy prepared for publication in the Federal Register is scheduled to appear on 19 November 2026. The rule says it will take effect based on its publication date. It also sets a deadline for public comments at 60 days after publication. The announcement on 16 November is a separate event and does not start this legal timetable.

Section 744.24(b)(3) gives data centers that were already providing covered computing services before publication 90 days from the rule's effective date to get a license if they want to continue those services. Research offices checking affected arrangements can use the published notice to set their schedules. They should keep Guo's diplomatic promise separate from the US rule's licensing requirements.

Sources