LIVE
EXPORT CONTROLS

US Commerce Department moves to ban Chinese AI labs from renting Nvidia Blackwell and Rubin chips via third-country clouds

An interim final rule drafted by the Bureau of Industry and Security closes the "Infrastructure-as-a-Service loophole," requiring cloud operators in Singapore, Malaysia, and the UAE to verify beneficial ownership before provisioning frontier GPU clusters.

UpdateNext Staff
Editorial team
5 min read
2 views
US Commerce Department closes offshore cloud GPU rental loophole for Chinese AI labs
Liquid-cooled AI GPU supercomputer clusters in Southeast Asia and the Middle East face mandatory beneficial-ownership KYC checks under new U.S. Commerce Department rules. Illustration: UpdateNext

For more than two years, the most glaring vulnerability in Washington’s semiconductor export control regime was not physical chip smuggling across borders — it was a fiber-optic cable. Under existing Bureau of Industry and Security (BIS) rules, while shipping an Nvidia B200 GPU to Beijing or Shenzhen was strictly illegal, a Chinese AI startup could legally incorporate a subsidiary in Singapore or Dubai and rent 16,000 of those exact same GPUs remotely over the cloud.

On Sunday, the U.S. Department of Commerce moved decisively to slam that door shut. According to regulatory filings circulated to cloud infrastructure providers and reviewed by UpdateNext, BIS will publish an Interim Final Rule this week extending Export Administration Regulations (EAR) restrictions to cover remote compute provisioning of advanced AI accelerators exceeding 10^26 floating-point operations per second of aggregate cluster interconnect.

ADVERTISEMENT
Get the daily AI briefingSubscribe

Hardware-Enforced "Know Your Customer" for AI Clouds

Under the new framework, U.S. hyperscalers and foreign neocloud operators that purchase U.S.-designed AI accelerators (including Nvidia’s H200, GB200, and B300, as well as AMD’s MI350 series) must obtain a specific BIS license before granting bare-metal or virtualized training access to any entity whose ultimate parent company is domiciled in Country Group D:5 (which includes China).

Impact on Neoclouds and Domestic Chinese Silicon

The rule represents a direct hit to regional GPU cloud operators in Southeast Asia that financed massive Blackwell buildouts over the past 18 months on the back of demand from offshore affiliates of Chinese technology giants. Industry analysts estimate that between 15% and 22% of regional neocloud GPU utilization in Malaysia and Singapore was ultimately funded by Chinese engineering teams training multimodal foundation models offshore.

For Chinese AI laboratories, the closure of the offshore rental valve leaves no alternative but rapid migration onto domestic silicon stacks led by Huawei’s Ascend 950 and Alibaba’s newly unveiled Zhenwu V900 processor.

✓ WE KNOW
  • The U.S. Bureau of Industry and Security (BIS) has finalized a sweeping rule extending export licensing requirements to remote cloud rental (IaaS) of Nvidia H200, Blackwell B200, and Vera Rubin GPUs by entities headquartered in China or Macau.
  • Data center operators in Tier-2 partner nations — including Singapore, Malaysia, Indonesia, and Gulf states — must implement hardware-level Know-Your-Customer (KYC) telemetry or lose their Validated End-User allocations.
  • The crackdown follows forensic disclosures linking offshore shell-company GPU rentals to both frontier Chinese model training runs and the Nvidia board investigation.
? STILL UNCLEAR
    “Renting a banned supercomputer over a fiber-optic cable from Singapore is functionally identical to shipping the racks to Shenzhen. The cloud loophole era is over.”
    —
    SOURCES:U.S. Bureau of Industry and Security (BIS): IaaS Know-Your-Customer & Advanced Compute Rulemaking ↗
    > how_we_reported_this
    • ✓Sources: written from primary sources and official statements, linked above.
    • ✓Fact-check: names, dates and claims checked against the original sources.
    • ✓AI use: researched with AI tools; written and reviewed by our editors. Our AI policy
    • ✓Independence: UpdateNext has no financial relationship with the companies covered.
    2 views
    > react:
    UpdateNext Staff
    The UpdateNext newsroom covers AI and tech from primary sources, every day. Every story is checked against official statements.
    Write for us →
    Comments
    Comments are moderated · Sign in to comment
    > read_next
    MEMORY WALL BREAKTHROUGH
    TSMC draws up plans to enter 3D DRAM stacked memory manufacturing at Nvidia’s request to break the data-center memory wall
    HBM4 SUPERCYCLE
    SK Hynix, Samsung, and Micron report HBM4 memory sold out into 2027 as custom 3nm logic base dies reshape GPU packaging
    FOUNDRY PRICE HIKES
    TSMC and OSAT partners hike 3nm, 2nm, and CoWoS advanced packaging quotes by 10% to 20% as AI accelerator demand triples
    What's next in AI.

    The daily briefing on AI news, tools and breakthroughs.

    US Commerce Department moves to ban Chinese AI labs from renting Nvidia Blackwell and Rubin chips via third-country clouds | UpdateNext