US Prepares Ban on Chinese AI Firms Accessing American Accelerators via Overseas Data Centers

The US Department of Commerce is finalizing a new rule that will prevent Chinese AI companies from renting computing power on American accelerators via data centers located outside the PRC. This is reported by The Information, citing two sources familiar with the rulemaking process. According to the source, the new restrictions could be officially announced as early as November 2026. The rule will be a logical extension of existing semiconductor export controls led by the Bureau of Industry and Security (BIS) at the US Department of Commerce and will significantly expand its perimeter to cover indirect chip access mechanisms.

The Circumvention Mechanism and Its Closure

Chinese tech firms turned to renting foreign data centers after Washington imposed direct restrictions on the shipment of advanced accelerators to the PRC. According to Reuters, Tencent, Alibaba, and ByteDance concluded agreements with data center operators in third countries, gaining access through them to powerful Nvidia chips required for training large language models. In effect, this allowed Chinese developers to use American semiconductors without formally violating the letter of export restrictions aimed at direct shipments. The Commerce Department's new rule closes this very channel: the ban will extend to remote access to computing resources located within sanctioned jurisdictions.

Geography of Restrictions: Seven Countries Targeted

The new restrictions will affect seven more countries through which, according to US regulators, Chinese developers were gaining access to foreign computing power and organizing accelerator supplies bypassing direct bans. Among the mentioned jurisdictions are Singapore, Malaysia, and Thailand. The choice of these states is no coincidence: all of them are major data center hubs in Southeast Asia and attract tech companies with tax incentives, developed infrastructure, and proximity to the Chinese market. Expanding the list of sanctioned territories means that data center operators in these countries will have to review service terms for PRC clients and implement additional end-user verification procedures.

Nvidia's Position and Industry Contradictions

Nvidia had opposed further tightening of export controls. The company insisted that excessive restrictions hinder American semiconductor manufacturers from competing globally and effectively cede a competitive advantage to Chinese developers, who in turn accelerate their own import substitution programs. Nvidia’s argument is that a ban on indirect access reduces the legal sales market for US chips without stopping the development of China’s domestic semiconductor industry. Nevertheless, the US Department of Commerce, based on current data, maintains a tougher position: the final version of the rules is being coordinated with other agencies at the time of writing, and there are no signs of significant softening in the text.

Impact on the Global AI Ecosystem

If the rule comes into effect as stated, it will have a direct impact on the architecture of training advanced AI models in China. Tencent, Alibaba, and ByteDance, which, according to Reuters, have already invested significant funds in foreign infrastructure, will be forced to either abandon part of their training clusters or repurpose them for domestic accelerators. For data center operators in Singapore, Malaysia, and Thailand, this means the risk of losing a large segment of revenue and the need for investment in compliance infrastructure. For the American semiconductor industry, on the contrary, reducing legal exports to workaround jurisdictions may accelerate the Chinese market's transition to alternative platforms, which in the long run will narrow the market share of Nvidia and AMD in Asia.

Timeline and Next Steps

At the moment, the final text of the rule has not been published. The US Commerce Department is conducting final interagency coordination, which usually takes from several weeks to two months. The Information indicates that the announcement could take place as early as November 2026. After publication in the Federal Register, stakeholders will have a period to submit comments, after which the rule will enter into force. For Chinese companies and data center operators in the affected jurisdictions, this creates a preparation window: reallocation of workloads, revision of contracts, and, in some cases, a transition to local computing resources.