TECH2030
Insights
Policy · 4 min readAI-assisted

US to Require Export Licenses for China's Remote Use of Overseas GPUs

US BIS now requires export licenses for overseas data centers providing over 12M TPP of computing resources to Chinese firms and others.

US to Require Export Licenses for China's Remote Use of Overseas GPUs

Key Takeaways

  • US approval required when providing over 12M TPP
  • Rules examine end users and purpose, not just sales destination
  • Domestic GPU rental providers should start by reviewing customer composition
  • Which country's GPUs is our AI workload running on?

On November 16, 2026 (local time), the US Department of Commerce's Bureau of Industry and Security (BIS) released an Interim Final Rule that brings even the remote use of advanced GPUs in overseas data centers by Chinese companies under export license requirements.

The US has long restricted direct sales of advanced AI chips to China, but Chinese companies have been able to remotely lease GPUs installed in other countries, such as those in Southeast Asia, to develop AI models. This new rule now subjects AI model training by Chinese companies using overseas data centers to export controls as well. In the same announcement, the US government also formally rescinded the 'AI Diffusion Rule' introduced by the Biden administration in 2025.

The Department of Commerce explained that the rule is "designed to prevent circumvention of existing restrictions on access to sensitive US technology through remote computing via third-country data centers."

What Changes Under the New Rule

  • Overseas data centers that provide more than 12 million Total Processing Performance (TPP) of computing resources to companies headquartered in countries under US arms embargo, including China and Russia, and in Macau, must obtain export licenses from the US government.
  • The scale is determined by aggregating the computing resources of multiple networked data centers, not just a single facility.
  • Using GPUs through third-country intermediaries or outsourcing AI model training to other companies is also subject to regulation.
  • Data center operators and semiconductor exporters must verify the actual ownership and control relationships of client companies as well as the ultimate purpose for which the computing resources will be used.
  • Overseas data centers already providing computing resources to Chinese companies and others must obtain a license within 90 days of the rule's effective date to continue their services.
  • Computing resources used solely for product and content recommendation systems on e-commerce platforms and social media are exempted if certain conditions are met.

How Exports to 5 Countries, Including Singapore and Malaysia, Will Change

Separately, the US government also tightened export license requirements for advanced AI chips to five countries: Singapore, Malaysia, Thailand, Indonesia, and Brazil. Exports or re-exports of advanced semiconductors to these five countries will now require additional licenses, with the US government reviewing approval on a case-by-case transaction basis. Chips supplied to companies headquartered in the US or whose ultimate parent company is a US firm are exempt from the additional licensing requirement if certain reporting conditions are met.

Implications for Korean Companies' AI Transformation (AX)

This rule places weight not only on which country a chip is sold to, but also on verifying who actually uses the computing resources and for what purpose. Korea was not included among the five countries subject to tightened export license requirements. However, companies operating data center or GPU rental businesses domestically should first check whether any of their clients are headquartered in arms-embargoed countries such as China or Russia, or in Macau, and whether the aggregated scale of networked resources exceeds 12 million TPP.

Companies training AI models using overseas cloud GPUs should check whether those GPUs are located in the Singapore, Malaysia, Thailand, Indonesia, or Brazil regions, and it would be worth asking providers whether there are any changes to capacity expansion or supply schedules. There is one thing worth checking this week: whether your organization has a list documenting which country and which provider's GPUs are running your AI training and inference workloads.

Source: US Blocks China's Workaround Use of Overseas GPUs... Announces New Export Regulations

If you found this helpful, share it.
TECH2030 service for this topicCloud & System Operations
Enterprise AX
Want to apply this to your own operations?
Request an AX consultation