Nvidia is reportedly considering a significant increase in the production of its H200 AI chips to meet a surge in demand from Chinese companies. This potential ramp-up follows the U.S. Department of Commerce's recent approval for Nvidia to export these powerful graphics processing units (GPUs) to China, Reuters reported, citing anonymous sources.
The H200 chips, representing the most potent iteration of Nvidia's Hopper generation GPUs designed for training large language models, were previously restricted from sale in China. The Biden administration had proposed extensive rules limiting the export of advanced AI chips to the country. However, the Department of Commerce recently granted Nvidia permission to sell H200 GPUs in China, reportedly in exchange for a 25% share of the sales revenue.
The reported surge in orders from Chinese firms has prompted Nvidia to explore expanding its production capacity for the H200. Despite this, Chinese officials are still deliberating on whether to permit the import of these chips, which are noted to be substantially more powerful than the H20 GPUs Nvidia had previously customized for the Chinese market.
For Nvidia, increasing H200 GPU production offers an opportunity to capitalize on significant untapped demand within China. The country is actively striving to develop its own homegrown AI chips, facing limitations on access to the most advanced AI training hardware due to Western competition and national security concerns. Consequently, Chinese companies have often prioritized efficiency in their AI models over raw computational scale.
Major Chinese technology companies, including Alibaba and ByteDance, both actively developing their own AI models, have reportedly initiated discussions with Nvidia regarding substantial orders for the H200 chips, which are currently produced in limited volumes.
"We are managing our supply chain to ensure that licensed sales of the H200 to authorized customers in China will have no impact on our ability to supply customers in the United States," an Nvidia spokesperson said in an emailed statement.
