Reuters: About 600,000 NVIDIA H200s Held in Hong Kong Awaiting Chinese Import Approval

On November 6, 2026, mlq.ai reported, citing Reuters, that about 600,000 NVIDIA H200s are in Hong Kong awaiting import approval from Chinese authorities. Reuters cited two people familiar with the matter.

The approximately 600,000 units cited in the Reuters report are said to be inventory in Hong Kong; they do not directly indicate the scale of computing resources available in mainland China. If import approval and actual shipment volumes remain separate, inventory figures alone cannot assess the supply of AI computing infrastructure within China.

The report focuses not on H200 production or shipping capacity, but on the lack of import approval for moving the approximately 600,000 units said to be in Hong Kong into mainland China. Even if the physical inventory exists, stock awaiting approval is not the same as computing resources operating on the mainland. Therefore, this figure cannot be treated as the number of GPUs available in China or as computing capacity that can be used immediately. Reuters reported the inventory figure based on information from sources, but the article provided does not explain how the inventory was verified or give individual shipment schedules. The “approximately 600,000 units” should be treated as a reported estimate, separate from any confirmed facts about the scope or timing of import approval. Reuters also reported shipment conditions and expenses based on NVIDIA’s 10-Q, which may indicate a gap between export licenses and actual shipments, but does not directly link those details to the entire Hong Kong inventory. From a physical AI perspective, access to computing semiconductors is one element of the infrastructure supporting robot training and inference. However, there is no information that these H200s are earmarked for robotics. The approximately 300 TB of open data discussed in yesterday’s [Hugging Face Releases LeRobot v0.7.0, a Robot Training Platform Supporting Conversational Control and Large-Scale Data Training](/en/news/hugging-face-releases-lerobot-v0-7-0-robotics-training-platform-adding-4e8d3f) concerns a training-data infrastructure and is separate from the import-approval issue. The availability of data does not determine whether computing resources can be used; that depends on separate factors. The points to watch are the basis for the reported inventory figure, the scope of any import approval and the quantity actually shipped and put into operation after approval. Without this information, the impact on computing capacity in China or on specific industrial applications cannot be quantified. Reuters’ report highlights approval as a constraint in the supply chain, but does not indicate the scale of computing resources ultimately put into operation.