Chinese military organizations, state-affiliated artificial intelligence research institutions, and universities have procured relatively small quantities of Nvidia semiconductors, despite the U.S. ban on their export to China.
H100 platform. Image credit: NVIDIA
According to a Reuters analysis, these purchases were conducted last year and were facilitated by relatively unknown Chinese suppliers. Such facts underscore the challenges faced by Washington in preventing China’s access to advanced U.S. chips, particularly those with applications in artificial intelligence and high-performance computing for military use.
Despite export restrictions, the publicly available tender documents reveal numerous instances of Chinese entities acquiring Nvidia semiconductors, including the A100, H100, A800, and H800 chips, after the bans were implemented.
Nvidia’s graphic processing units (GPUs) are widely acknowledged for their superior performance in AI tasks, efficiently handling large datasets required for machine learning. The persistent demand for and access to banned Nvidia chips is driven by the limited alternatives available to Chinese firms, despite emerging competition from companies like Huawei.
Nvidia previously dominated China’s AI chip market with a 90% share before the imposition of export restrictions.
Buyers include prestigious universities and entities, such as the Harbin Institute of Technology and the University of Electronic Science and Technology of China, both subject to U.S. export restrictions. Local vendors reportedly acquire excess stock from large U.S. firms or import through companies in regions like India, Taiwan, and Singapore.
The Reuters analysis covers over 100 tenders where state entities acquired A100 chips, and post-October bans reveal A800 purchases in dozens of tenders. Documents indicate Tsinghua University procured two H100 chips, and a lab under the Ministry of Industry and Information Technology obtained one. A military entity in Wuxi sought three A100 chips in October and one H100 chip in the current month.
Although military tenders are typically redacted, most indicate AI usage. While the quantities are small, they can perform complex machine-learning tasks and enhance existing AI models.
Written by Alius Noreika