By Kate Yuan
(JW Insights) Oct 19 -- The upgraded US curbs on AI chip exports will accelerate the comprehensive breakthrough of the Chinese AI chip industry chain, said a commentary by JW Insights analyst Li Ying.
The US Department of Commerce issued new regulations regarding the export ban on AI chips on October 17. The upgraded rules are designed to prevent China from acquiring the cutting-edge chips needed to develop AI technologies such as large language models, indicating the US’ comprehensive curbs of China’s AI chips.
The rules introduced on October 7, 2022 mainly focused on computing power and interconnect bandwidth, with a computing power limit of 4800 and a bandwidth limit of 600 GB/s. This new ban removed the parameter of "interconnect bandwidth" and imposed a "performance density" limitation instead.
Not only NVIDIA, but companies like Intel and AMD also have virtually no way to supply latest products to China. Products from the previous two or three generations can be exported at most. Chinese AI chipmakers will also be unable to produce mainstream products at foundries with advanced process technology such as TSMC.
In addition, the US has imposed additional license requirements on products exported to more than 40 countries. This means that all Chinese companies, whether located within China or overseas, will find it difficult to obtain chips subject to US export controls.
The upgraded ban by the US aims not only to strangle China’s high-end computing power chip industry but also to block China's path to advancing in the AI revolution.
In the short term, Chinese AI chips face challenges such as increased power consumption and reduced energy efficiency due to the inability to use advanced EDA tools and IP from overseas and advanced process.
In the long run, domestic AI chip companies must rise to the challenge and build a sound and independent supply chain that is not subject to US restrictions.
Experts suggest that efforts should be made to strengthen weaknesses in EDA tools and IP. On the other hand, it is important to build advanced process manufacturing lines, overcoming challenges in chip fabrication.
At the same time, barriers between industry applications, chip development, system development, and university research should be broken through, forming cross-enterprise, cross-domain, and cross-industry cooperation to promote the all-dimensional development of the chip industry.
In recent years, AI and GPU companies in China are making progress against the backdrop of escalating sanctions and a shortage of computing power. Huawei, Alibaba, Baidu, and Tencent have all developed their own AI chips, and AI chip companies have achieved fruitful results in architecture innovation, design, foundry, and applications.
Data shows that Chinese AI chipmakers are experiencing rapid growth. In the first half of 2023, the market size of China's accelerated chips exceeded 500,000 units, with GPUs accounting for 90% of the market share. Chinese AI chip brands have shipped over 50,000 units, accounting for about 10% of the entire market share, according to the JW Insights reports.
