In October of this year, the US Department of Commerce announced a new package of sanctions that tightened restrictions on the export of high-performance chips and the tools used in their manufacture. The intention is to prevent Chinese companies from sourcing technology equipment from the US through foreign subsidiaries and to prevent Beijing from using artificial intelligence for military purposes, which is a major concern for the US government. In response, China has accused the US of trying to impede its technological and economic progress, framing this as an economic conflict.
The current sanctions package tightens restrictions originally introduced by the US last year, which marked the beginning of intensified “technological competition” between the two countries. They were intended to restrict China’s procurement of powerful training chips capable of high-speed data exchange, which serve as a key element in assembling powerful computer networks and for the development of artificial intelligence. However, after these measures were put in place, certain legal and technical shortcomings became apparent over time, allowing Chinese companies to continue procuring technology components.
The restrictions last year particularly affected large manufacturers such as Nvidia, which will no longer be able to sell Beijing H100 and A100 chips, which are considered the most powerful AI training chips. The US tech giant immediately responded to this by developing suitable alternatives called H800 and A800, which effectively bypassed previous regulatory mechanisms. For both new products, the communication with other chips is at a slightly lower communication speed, albeit only slightly below the threshold set by previous regulations. Despite the performance limitations compared to the original chips, the H800 and A800 have proven valuable for the development of AI applications. This was obviously a successful business move on Nvidia’s part. Major Chinese tech companies Baidu, Alibaba, Tencent, and Bytedance reportedly placed billion-dollar orders for H800 chips in recent months. However, it is uncertain whether these sales will be completed before the new restrictions take effect.
Beijing’s significant progress in AI development is also evidenced by the unveiling of Ernie 4.0, the most advanced language model yet created by Baidu. The company claims that this model will match the AI model behind ChatGPT in performance. Although the Chinese company did not specify the type of chip used, an anonymous source confirmed that chips from the aforementioned American concern were involved in the training process.
However, in general, the restrictions have caused quite a stir among US chipmakers, with some expressing concern about the possibility of further regulations. The Semiconductor Industry Association, the representative of US chip companies, has also expressed concerns. While acknowledging the need to protect national security, they argue that unilateral and disproportionately broad regulations could threaten the industry in the US and force foreign buyers to look elsewhere in the world for products. They therefore call on the government to ensure fair competition for all businesses.
Gregory Allen, Director of the Centre for Advanced Studies at the Centre for Strategic and International Studies, also supports the new restrictions and stresses their importance. In his view, restrictions on chip-making facilities could be just as important as stricter rules on AI training chips. These new rules will prohibit the sale of certain devices, moving away from the previous approach of focusing on end-use and preventing Chinese companies from hiding their true intentions regarding the devices’ usefulness. U.S. Commerce Secretary Gina Raimondo said that these controls remain clearly focused on military applications and address the threats to national security posed by the PRC government’s military-civilian fusion strategy.
The above-mentioned US attitude to the whole issue was further illustrated by two recent events. On 13 October, Washington announced that it would allow South Korean companies SK Hynix and Samsung to continue to receive some US chip-making tooling in their Chinese plants. This decision will help increase the capacity of South Korean semiconductor manufacturers in China. This is in line with the US strategy to support foreign companies operating in the PRC. On the same day, Taiwanese company TSMC announced that it would receive a permanent authorisation to operate its Chinese chip manufacturing plant in Nanjing. TSMC is the largest semiconductor chip manufacturer in the world. The decision is significant because TSMC had previously only received an extension of a waiver that allowed it to use U.S. facilities to manufacture chips in China.
Thus, it can be concluded that the US government’s decision to tighten sanctions on chip exports reflects its desire to further restrict China’s access to cutting-edge technologies with potential military applications, particularly in the field of artificial intelligence. Although these regulations are intended to restrict the sale of chips and equipment, the US semiconductor industry is divided and concerned about their broad reach. China’s response to these restrictions could be to find new ‘creative’ ways to circumvent them. In fact, such action could enhance the competitiveness of China’s semiconductor industry, which has the ambition to build up its computing power capacity to 300 exaflops by 2025 (1 exaflop is equivalent to the power of two million laptops). This ongoing technological rivalry between the US and the PRC has significant implications for both countries as well as for the global technology sector.
This brief is supported by
NATO’s Public Diplomacy Division