The development of China’s artificial intelligence sector is expected to be slowed in the coming few years by the United States’ new ban on exports of several high-end chips made by Nvidia and AMD, say Chinese IT experts.
Nvidia said last Friday it had been informed by the US government that it must stop exporting its graphics processing unit (GPU) chips, namely A100 and H100, to China and Russia. It said its DGX, an AI server, was also barred from being shipped to China if a unit contained the two chips.
At the same time, media reports said the US had also restricted sales of AMD’s MI250 Accelerator AI chip to China.
China’s Foreign Ministry said the US had typically exerted its “sci-tech hegemony” and violated the rules of the market economy with its latest chip-export restrictions.
Most IT columnists and experts in mainland China said the ban would deal a heavy blow to China’s AI sector, which currently relies on Nvidia and AMD chips. A few said that, due to the ban, Chinese IT giants would boost expenses in research and development to make their own AI chips.
Last October, US Trade Representative Katherine Tai said the US would “deploy all tools and explore the development of new ones, including through collaboration with other economies and countries,” to address fundamental concerns related to China’s trade practices.
Tai said China had already dominated the world’s steel and solar cell production through its “unfair” policies and was planning to do the same in the semiconductor industry by 2030.
She said the US had to take a new, holistic and pragmatic approach in its relationship with China.
Last month, the US Department of Commerce’s Bureau of Industry and Security banned the export to China of electronic computer-aided design (ECAD) software that is used for designing 3-nanometer (nm) chips.
Handel Jones, founder and chief executive of International Business Strategies (IBS), wrote in a report that the ban would have an impact on the trade between China and the US greater than the combined impact of all other actions taken to date.
Jones said the limitation on software affected certain EDA tools for gate-all-around, a transistor structure, including the 2nm technology of TSMC and Intel as well as the 3nm and 2nm technologies of Samsung Electronics.
He added that the restriction would also hurt US companies as they would lose sales in the China market.
In early 2020, Washington had already stopped the Netherlands from exporting its extreme ultraviolet (EUV) lithography equipment to China, meaning that China could not mass produce high-end chips between 22 nm and 7nm since then.
Hong Shibin, an IT industry expert, wrote in an article on August 15: “The US wants China to be stopped at 5-nm in chip design and 7-nm in chip manufacturing so China’s growth in high-speed computing and AI will slow.”
On August 9, US President Joe Biden signed the CHIPS and Science Act to boost domestic semiconductor production and scientific research to enhance US competitiveness vis-a-vis China. The new law grants US$54 billion of subsidies and tax benefits to US-based chip makers over the next five years.
Last week, when the US government ordered Nvidia and AMD to stop exporting their GPU and AI chips to China and Russia, it explained to Nvidia that its A100 and H100 chips could be used in, or diverted to, a “military end use” or “military end user” in the two countries. Nvidia said it did not sell products to customers in Russia.
Chinese IT columnist Li Wei said China’s internet giants would no longer be able to obtain the most cutting-edge chips from Nvidia and AMD, meaning that their developments in high-speed computing and AI would be slowed. Li said it would be difficult for these companies to maintain their advantages in the global AI sector.
In the three months ended July 31 this year, Nvidia’s revenue from the data center segment jumped 61% to $3.81 billion from a year ago. That accounted for 57% of the company’s total revenue.
Statistic.com said Nvidia generated 26.4% of its revenue from China in the fiscal year ended January 30, compared with 31.7% from Taiwan and 16.2% from the US.
Nvidia has a 80.6% share in the global market for artificial intelligence processors used in the cloud and in data centers in 2020, according to Omdia, a data service provider.
In recent years, many AI chip makers have been emerging in China domestically. Cambrian Technologies Corp and Changsha Jingjia Microelectronics Co Ltd have been listed in Shanghai and Shenzhen, respectively. Newcomers include Biren Technology, Moore Threads, Iluvatar CoreX and Shanghai Enflame Technology.
Zhao Lidong, chief executive and founder of Shanghai Enflame Technology, said at the World Artificial Intelligence Conference in Shanghai on September 1 that Chinese chip makers were lagging behind their counterparts in the US and Europe, which had spent decades in research and development.
However, Zhao added that he believed Chinese chip-design firms would succeed one day if they continued to innovate over the long run.
Guo Yiwu, Secretary-General of Shanghai Integrated Circuit Industry Association, said it would take a very long time for Chinese GPU makers to catch up with Nvidia. Guo said Chinese chips could only compete with the international ones in a few areas.
Chinese media reports said Cambrian, established in 2016, had been a rising star in the industry a few years ago as its neural processing unit (NPU) module was used as a part of Huawei Kirin 970 in 2017.
They said Cambrian had burnt more than 2 billion yuan ($288 million) in R&D between 2017 and 2020 but then the company had faced a widening loss after its major customer Huawei Technologies shifted to use its own Da Vinci AI chip architecture in 2020.
An IT columnist said that what Chinese chip makers lack is not the technology needed to design high-end chipsets but, rather, an ecosystem that could support their growth.
Since Chinese internet giants now cannot obtain the most advanced GPUs from the US, he said, they will boost their investments to make their own AI chips. He said Baidu’s Kunlun, Alibaba’s RISC-V and Tencent’s Zixiao AI chips are catching up.
Follow Jeff Pao on Twitter at @jeffpao3