ByteDance Leads Chinese Companies in Nvidia Chip Purchases This Year
Partial Approval for Training Chips
Pressure to Use Chinese Chips for Inference Tasks
The Chinese government has reportedly demanded that ByteDance, the parent company of TikTok, halt the use of Nvidia chips. However, the government is still partially allowing the use of Nvidia chips for AI model training. Analysts interpret this as a move by the Chinese government to reduce its dependence on the United States in the AI sector and to foster its domestic semiconductor ecosystem.
The corporate logo is displayed on the exterior wall of ByteDance's headquarters in Shanghai, China. Photo by Reuters Yonhap News
According to a report by The Information, a U.S. IT-focused media outlet, on the 26th (local time), Chinese regulatory authorities have blocked ByteDance from using Nvidia chips in its new data centers, citing sources familiar with the matter.
The authorities have not only prohibited new orders of Nvidia chips but have also instructed the company not to use chips it already possesses. This measure is even stricter than China's previous directive, which only banned local companies from placing new orders for Nvidia's AI chips. ByteDance is known to have been the largest purchaser of Nvidia chips among Chinese companies this year.
This move is seen as a reflection of China's strategic intent to reduce reliance on the U.S. AI ecosystem and to nurture domestic technology. Ultimately, it is interpreted as pressure to use products from Chinese companies such as Huawei or Cambricon instead of Nvidia chips.
Brady Wang, an analyst at Counterpoint Research, explained that Chinese companies are moving quickly to reduce their reliance on Nvidia chips, saying, "This is not something they want to do, but something they have no other choice but to do."
Sources say that China has only banned the use of Nvidia chips for running and inference tasks of AI models, and has not completely prohibited their purchase for model training purposes. This is because Chinese-made chips are technically capable of handling AI model operations, but still fall short when it comes to training AI models. China also lacks sufficient production capacity for AI chips.
Currently, the United States only permits the sale of Nvidia's lower-performance 'H20' chips to China. However, these chips are considered more suitable for running and inference tasks than for training. On the other hand, if the U.S. allows the export of the 'H200'-which is currently under review-the situation could change, as the H200 offers about twice the performance of the H20 and is also suitable for model training.
© The Asia Business Daily(www.asiae.co.kr). All rights reserved.
