AI chip ban update 2.0 domestic companies seek countermeasures

Chips have become the focus of Sino-US friction. On October 17, the Biden administration updated export control regulations for artificial intelligence chips, planning to prevent companies such as NVIDIA from exporting advanced AI chips to specific countries. On the same day, the Bureau of Industry and Security (BIS) under the US Department of Commerce added 13 Chinese GPU companies to the entity list, including Moore Thread and Wall Technology. The global artificial intelligence industry is in a critical competition period, and chips have become the "back-pot man" of politicization of economic and trade technology issues, but while the global supply chain is shaking, domestic AI chips may also accelerate new breakthroughs under pressure.

A800, H800 are listed

Last October, the United States imposed bandwidth rate limits on AI chips exported to China, involving NVIDIA A100 and H100 chips. At that time, in order to bypass restrictions, NVIDIA launched alternative versions of A800 and H800 for the Chinese market, which also became the "target" of the US chip export control.

According to the latest rules, NVIDIA's chip exports to China, including A800 and H800, will be affected, and the new rules will take effect after 30 days of public consultation. Commerce Secretary Gina Raimondo said the new measures were intended to "close loopholes" and said they could be updated at least annually in the future.

The chip is the watch, AI is the inside. Raimondo bluntly said the goal of the U.S. restrictions is to prevent China from acquiring advanced semiconductors that could drive breakthroughs in artificial intelligence, especially for military use. According to reports, the new restrictions will only affect a small part of chip exports to China, and chips used in consumer products such as game consoles or smartphones will not be subject to export controls.

On October 18, a spokesperson for China's Ministry of Commerce said that the United States continues to generalize the concept of national security, abuse export control measures, and commit unilateral bullying. China will take all necessary measures to resolutely safeguard its legitimate rights and interests.

It is understood that in the new regulations, the US Department of Commerce also tried to prevent chips from being shipped to China through other countries, extending the coverage of export restrictions to overseas subsidiaries of Chinese companies and 21 other countries and regions. In addition to Nvidia, the new rules could discourage companies such as ASML, Applied Materials, Lam Research and KLA from selling and exporting semiconductor manufacturing equipment to China.

NVIDIA responded: "We comply with all applicable regulations while striving to support thousands of applications in different industries. Given global demand for our products, we do not expect to have a material impact on our financial results in the near term. ”

On the evening of October 17, in response to being included in the Entity List, Moore Thread and Bicheng Technology both issued statements through their official public accounts, expressing strong opposition and protest to the US Department of Commerce's move, strictly abiding by the laws and regulations of relevant countries and regions, and always operating legally and in accordance with laws and regulations on this basis, while assessing the possible impact of the incident.

Liang Zhenpeng, a senior industrial economic observer, told the Beijing Business Daily reporter that the tightening of AI chip exports in the United States will have a certain impact on the development of the mainland's artificial intelligence industry and large models. First, Chinese companies may face challenges in technology supply and R&D capabilities because they cannot directly access or collaborate on advanced chips designed by U.S. companies. This may lead to the postponement or limitation of some projects, which will hinder the research and application of Continental in the field of large models.

"At the same time, because U.S. companies have a competitive advantage in chip technology, Chinese companies may be limited in the advanced chip market, reducing their competitiveness." Liang Zhenpeng said.

NVIDIA "lost" the Chinese market?

As soon as the "chip ban" update came out, NVIDIA's stock price immediately fluctuated sharply, once fell nearly 7%, as of the close of the day, NVIDIA's stock price fell 4.68%, and the market value evaporated more than 53.5 billion US dollars overnight, equivalent to nearly 400 billion yuan.

Most of the chip companies in the United States were also implicated, Intel closed down 1.37% on the day, AMD closed down 1.24%, Broadcom closed down 2.01%, TSMC, ASML, Qualcomm and other chip giants also fell to varying degrees.

The spread of panic is understandable. It is understood that NVIDIA is the "hegemon" of the GPU market. Data show that its global discrete graphics card market share is as high as 80%, high-end GPUs such as H100, A100 and V100, etc., occupy the vast majority of the AI algorithm training market.

The rise of large models and another wave of artificial intelligence have made GPUs suddenly hot, and NVIDIA has made a lot of money. NVIDIA's financial report for the second fiscal quarter of fiscal 2024 ended July 30 showed that revenue for the quarter was US$13.507 billion, an increase of 101% year-on-year and 88% month-on-month, setting a new record in history; Net income was US$6,188 million, up 843% year-over-year and 203% sequentially.

Among them, in the "data center" business, NVIDIA's revenue increased to 10.32 billion US dollars, which also set a historical record. Nvidia Chief Financial Officer Colette Curris mentioned in a conference call after the earnings report that most of the demand for artificial intelligence chips still comes from China, and Chinese buyer investment accounts for 20% to 25% of the total revenue of data center products.

"If we are deprived of the Chinese market, we will have no emergency measures, and there is no other China in the world." In May this year, NVIDIA CEO Jensen Huang mentioned that China accounts for about one-third of the US technology industry market, and as a source of components and end markets for its products, China cannot be replaced. For the corresponding preparation measures, possible impact and other issues, the Beijing Business Daily reporter contacted NVIDIA, but did not receive a reply as of press time.

Earlier this month, the US media also disclosed that the three major US chip giants Intel, Qualcomm and NVIDIA joined forces to resist the Biden administration's tightening of chip exports to China. The report mentioned that the Chinese market accounts for about one-third of the global semiconductor market, generating a total of more than $50 billion in annual revenue for NVIDIA, Intel and Qualcomm.

How far is there to replace with domestic production

Contrary to the reaction of U.S. chip stocks, on October 18, A-share computing power chip concept stocks generally rose, Holly Technology rose to the limit, Cambrian and Hongxin Electronics rose more than 10% intraday, and Jingjiawei and Haiguang Information closed up.

"If China can't buy it from the U.S., they'll make it themselves." The United States must be careful that China is a very important market for technology industries. In an interview with the Financial Times in May, Huang made such a prediction.

Guo Tao, an angel investor and senior artificial intelligence expert, told the Beijing Business Daily reporter that the impact of the tightening of AI chip exports in the United States on the mainland's artificial intelligence industry and the development of large models may be manifested as short-term supply shortages and rising costs, but the impact is relatively limited. In the long run, it is conducive to the healthy development of China's chip industry, and prompts chip companies to achieve the goal of changing lanes and overtaking through independent technological innovation.

Taking Huawei as an example, public information shows that Huawei's Ascend chips include the Ascend 310 and the Ascend 910, the former has an integer precision computing power of up to 16TOPS, which is mainly used in low-power fields such as edge computing products and mobile devices. The latter's integer precision computing power can reach 640TOPS, and the performance level is close to that of NVIDIA A100.

Last month, Meng Wanzhou, vice chairman, rotating chairman and CFO of Huawei, also mentioned in a speech that Huawei will build a solid computing power foundation in China and build a second choice for the world. On the same day, Huawei released a new generation of Ascend computing products, and launched the Atlas 900 SuperCluster, a new architecture of Ascend AI computing cluster, which can support large-scale model training with more than one trillion parameters.

IDC's latest report points out that with government support, China's local cloud AI-accelerated chip manufacturers are also growing rapidly. The latest data shows that in the first half of 2023, 500,000 locally sourced/developed AI accelerator chips were used in China's AI servers. This volume accounts for 10% of the entire server market in China.

Guo Tao mentioned that in recent years, the mainland AI chip industry has developed rapidly, making new breakthroughs in a number of key core technology fields, and now has the ability to produce chips comparable to NVIDIA A100, but if you want to achieve large-scale shipments, you still face many challenges.

"At present, China's semiconductor chip industry is still facing some challenges, including technology research and development, production process, equipment manufacturing and other issues. In addition, the shortage of talents is also an important factor restricting the development of China's semiconductor chip industry. Guo Tao said.

Liang Zhenpeng also mentioned that the Chinese government and related enterprises are increasing support and investment in the semiconductor chip industry and striving to promote the development of the industry. New export controls may accelerate the development of domestically produced AI chips, as Chinese companies will be forced to find alternative suppliers or develop alternative products on their own. This will help promote the improvement of domestic chip industry technology and innovation capabilities, gradually reduce dependence on imported technology, and promote the localization process.

"However, it will take time and effort to achieve a complete domestic substitution." Liang Zhenpeng said that domestic enterprises need to further increase efforts in research and development investment, technical reserves, and talent training. In addition, China's semiconductor chip industry is still facing bottlenecks in some key technologies, such as process technology and chip design capabilities.

Beijing Business Daily reporter Yang Yuehan