Overseas access to NVIDIA AI accelerators is the focus of a US export-control change reported by TechPowerUp. Its November 17 report says the Bureau of Industry and Security released an interim final rule on November 16 addressing companies that rent computing capacity abroad instead of buying restricted chips directly.
According to the report, shipments of advanced accelerators to data centers anywhere in the world require a license when those facilities provide more than 12 million aggregate Total Processing Performance, or TPP, to companies headquartered in Macau or Country Group D:5, which includes China.
Cloud access falls within the scope
TPP measures chip computing power using peak dense throughput and the operation’s bit length. TechPowerUp puts the threshold at approximately 750 NVIDIA H100 accelerators or 330 B200s. The reported restrictions also cover linked data centers and indirect arrangements, including a cloud reseller training models for a Chinese company.
Existing facilities serving affected customers have 90 days to obtain a license, the report says. An exemption covers commercial recommender systems, but excludes systems generating text, images, video or code.
Additional destinations face licensing requirements
The report also identifies Indonesia, Malaysia, Singapore, Thailand and Brazil as destinations requiring licenses for advanced AI chip shipments. It attributes that measure to Commerce Department concerns about diversion. Shipments to US-headquartered companies are exempt when new reporting requirements are met.
Separately, TechPowerUp reports that the rule formally rescinds the Biden-era AI Diffusion Rule, removing its country tiers, Data Center VEU program and model-weight controls.
For readers following NVIDIA and PC technology, the distinction is that this report concerns advanced AI accelerators and access to data-center computing. It provides no consumer graphics-card pricing or gaming-performance findings.
