Table of Contents
The US government has imposed strict export restrictions on Nvidia, the leading supplier of artificial intelligence (AI) semiconductors in China. This has prompted Chinese AI companies to turn to domestic chipmakers for their needs. However, the quality and quantity of Chinese AI chips remain uncertain, and the gap with Nvidia may widen in the future.
Nvidia faces headwinds in China
Nvidia, a US-based company, has dominated the Chinese AI chips semiconductor market for a long time, holding about 90% of the share. Its products, such as the H100 AI graphics processing unit (GPU), are widely used by Chinese tech giants like Alibaba, Tencent, and Baidu to power their servers and provide generative AI services.
However, in fall 2022, the US Commerce Department imposed export restrictions on Nvidia’s advanced chips, citing national security concerns. The restrictions aimed to prevent the chips from being diverted for military purposes by China or other countries.
Nvidia tried to circumvent the restrictions by shipping its lower-performance H800 chips to China, but these were also banned by the US in October 2023. As a result, Nvidia’s supply to China was severely disrupted, leaving many Chinese AI companies scrambling to find alternative sources.
Nvidia’s CEO Jensen Huang tried to appease his Chinese employees by sending them a video of him wishing them a happy Lunar New Year in Chinese and dancing in traditional Chinese clothing, as reported by the state-run Xinhua News Agency in late January. However, this gesture did not change the fact that Nvidia was facing strong headwinds in China.
Chinese AI chips emerge as alternatives
As the US is expected to continue tightening its export controls on Nvidia and other foreign chipmakers, many Chinese AI companies have started to switch to domestic products. Tencent’s president Martin Lau said in an earnings call in November that the company had enough stock of Nvidia’s chips for the short term, but it would also look for domestic sources for these training chips in the future.
One of the potential domestic suppliers is Huawei, which has developed its own AI chipset called Ascend. Despite being under US sanctions itself, Huawei managed to release a smartphone in August equipped with advanced 7-nanometer chips that it co-developed with Chinese contract chipmaker Semiconductor Manufacturing International Corp. (SMIC).
Huawei is also marketing its Ascend chips and related equipment and services to external parties. An executive at a major Chinese internet company said that Huawei’s chips were inferior to Nvidia’s, but they were usable.
Other Chinese chipmakers are also trying to take advantage of the opportunity created by Nvidia’s restrictions. Hygon Information Technology, a chip startup, launched its new AI semiconductor, the MTT S4000, at the end of the year. It claimed that its chip was compatible with Nvidia’s products and could easily copy data from them.
Hygon also announced that it was building AI infrastructure with Baidu and Alibaba, two of China’s largest internet companies.
Challenges remain for Chinese AI chips
However, the development of Chinese AI chips is not without challenges. The US has also targeted many Chinese chipmakers, including Hygon, with export restrictions. These companies do not have their own fabrication plants and rely on contract manufacturers like Taiwan Semiconductor Manufacturing Co. (TSMC), the world leader in chipmaking. However, the US has also restricted TSMC and other foreign chipmakers from supplying advanced chips to Chinese companies.
Huawei is also facing difficulties in mass production. Its Ascend chips are similar in performance to Nvidia’s for now, but they may fall behind in the future due to issues with their manufacturing process, according to Charlie Chai, an analyst at Shanghai-based 86Research. Some reports indicate that the yields for Ascend chips are as low as 20%, meaning that only one in five chips is functional.
In January, a Chinese government-affiliated research company projected that China’s global share of the AI industry would exceed 30% in 2035. However, this goal may be hard to achieve if the domestic chip industry fails to catch up with the US and other countries. China’s AI capabilities may be limited by the quality and quantity of its chips.
