China Will Allow Major AI Companies to Purchase Nvidia H200 Chips — The Information

China will for the first time in several years allow major AI companies to purchase Nvidia H200 chips — albeit in limited quantities and with government approval. This is reported by The Information citing sources familiar with the matter; the news is relayed by Bloomberg Tech.

What is the Nvidia H200 chip

The H200 is one of Nvidia's most powerful accelerators for training and inference of large language models, succeeding the H100 chip. Its main distinguishing feature is the increased HBM3e memory size of 141 GB and significantly higher memory bandwidth compared to its predecessor, which is critical when working with modern LLMs with billions of parameters.

• H200 is a development of the Hopper architecture, a key generation of Nvidia AI chips
• HBM3e memory at 141 GB represents a significant leap from the 80 GB of the H100
• The chip falls into the category of products subject to US export controls
• Only "top-tier" Chinese AI companies will be able to purchase it, and only in limited quantities

Why China was restricted from buying such chips

Since October 2022, the US Department of Commerce has restricted exports of Nvidia's most powerful AI accelerators — including A100, H100, and H200 — to China, citing risks of use in military and intelligence operations. Nvidia was forced to develop stripped-down versions of chips specifically for the Chinese market, including the H20 model, access to which has also periodically opened and closed against the backdrop of trade negotiations between Washington and Beijing.

What this means

If The Information's report is confirmed, this would be one of the most significant signals of an easing of US-China technological confrontation in AI computing: Chinese companies would gain access to more powerful hardware for model training, and Nvidia would regain a portion of the market lost due to sanctions.