The United States restricts high -performance semiconductor exports, which areas will be affected?

Economic Observation Network reporter Shen Yiran on August 31, the U.S. government ordered chip manufacturers Nvidia (NVIDIA) to stop selling some high -performance GPUs to China. According to Reuters, another super prestige semiconductor (AMD) also said that the relevant ban order was received. Essence

On September 1, a spokesman for the Ministry of Commerce, Shu Jingsing, said that China noticed the relevant situation. For a period of time, the United States has continuously abused export control measures to restrict the export of semiconductor -related items to China, and China has resolutely opposed it. The relevant practices of the United States depart from the principle of fair competition, violate the rules of international economic and trade, not only harm the legitimate rights and interests of Chinese enterprises, but also seriously affect the interests of American companies, hinder international scientific and technological exchanges and economic and trade cooperation, and cause the global industrial chain supply chain to stabilize and the world economy recovery Intellectual. The United States should immediately stop the error and treat enterprises including Chinese companies fairly, and do more things that are conducive to the stability of the world economy.

On August 31, Nvidia issued an announcement saying that as early as August 26 this year, the US government notified Nvidia that the company has implemented a new license require ,Effective immediately. Nvidia's A100 is a 4NMGPU chip, which is used in 2020; H100 is a 4nmgpu chip and launched in March this year.

Reuters reports that a spokesman for Chaowei Semiconductor told Reuters that it has received a new license request that will prevent its MI250 (6nm process, high -precision computing chip) chip export to China It will be affected.

Zhu Jing, deputy secretary -general of the Beijing Semiconductor Industry Association, said that according to the news, the blocked products are high -end GPUs with sufficient dual -accuracy computing power. As far as Nvidia is concerned, it is based on A100's computing power. Limited.

Nvidia announced that the new license requirement may affect the company's ability to complete H100 development or support existing customers in A100 in a timely manner. In addition, the company is in contact with Chinese customers and seek to meet their plans or buy the company's data center products in the future.

Nvidia said that if customers need new licenses to cover products, the company can apply for permits for customers, but it does not guarantee that the US government will grant any exemption or license for any customer, or the US government will take action in time.

The reporter further inquired with Chaowei Semiconductor and Nvidia, and as of the press time, it had not been responded. The stock price of Chaowei Semiconductor and Nvidia fell 2.38%and 2.42%respectively.

Zhu Jing said that this time the United States should be curbed for advanced computing (including supercomputing and intelligent computing), and the East Digital Western Calculation Strategy that China is starting will be temporarily affected because it requires computing power infrastructure support. In addition, it will temporarily expand to the Internet field.

High -performance GPU limit

The GPU is the graphics processor, which belongs to the field of logic device with a relatively complex and technical level requirements in integrated circuits. It is mainly used in PC graphics cards and AI computing. GMPU is a low self -sufficiency in domestic semiconductor, and China starts late.

Nvidia and Chaowei Semiconductor are the oligarchs of the GPU market. According to the JPR data of the consulting agency, in the PC GPU market in Q2 in 2022, the two companies occupy 18%and 20%, and the other accounts for 62%of Intel.

China is a strategic market for two companies. In the past year, Nvidia and Chaowei Semiconductor received US $ 7.111 billion and $ 4.096 billion from China. In Q3 this year, the potential sales of China's business affected by the new license are about $ 400 million.

Nvidia announced that the new license will not only affect the company's ability to develop H100 and sell A100 in a timely manner, but also may cause the company to transfer certain businesses outside China. The company predicts that this year's Q3, the potential sales of China's business affected by the new license will be about $ 400 million.

8月31日晚有市场消息称：“英伟达中国区已收到总部要求：暂停对中国区所有客户所有代理商的数据中心GPU卡A100和H100的发货，其它GPU卡不受影响；各服务器OEM's existing inventory A100 GPU card can be delivered to their respective industry customers. Nvidia China does not send any Letter on OEM at present; Nvidia Headquarters is still analyzing the policy requirements of the US government. It is expected There is a communication caliber of Chinese customers and partners. "

Rumors also include: Nvidia notify China's Internet client high -end AL training GPU chip, H100, A100 restricted sales of Chinese Internet companies. At present, Tencent, Baidu, bytes and other companies have verified that they have received notifications. There may be a clear news of sanctions, which are currently pre -communication.

According to financial reports, Tencent Cloud, Alibaba Cloud, Inspur, Lenovo, etc. are Nvidia customers. The reporter asked Alibaba, Tencent Cloud, and Inspur to verify, and as of press time, the other party did not respond.

There are also market sources that Chaowei Semiconductor informed that it received an emergency notice from the headquarters: 1. 2. Statistical China ML 100 has been shipped; Statistical statistics of the list of customers in China MI 200 and the delivery details super prestige semiconductor analysis may be that the U.S. government must restrict the sales of high -performance GPU cards in China, especially for China's HPC dual -precision and high -performance card. Pointing to China

Zhu Jing said that according to the news, the blocked product is a high -end GPU with sufficient dual -accuracy computing power, which is not affected to the low -end GPU. High -end GPUs with high dual -accuracy computing power are mainly used in high -performance computing fields, including scientific computing, CAE (computer -aided engineering), medical care, etc.

The supercomputing center is the National Super Computing Center. It consists of thousands of or more processors and has ultra -high computing power. It is known as "Mount Everest in the computer", which mainly meets the needs of national high -tech fields and cutting -edge technologies. There are supercomputing centers in Wuxi, Tianjin, Jinan, Shenzhen, Changsha, and Guangzhou in China.

In contrast, ordinary data centers are facing all scenarios that need information technology, including a large number of Internet applications. China's telecommunications operators and Internet companies have built data centers. Zhu Jing said that enterprise -level data centers often purchase A100 and H100 products in the news of Nivise. These products are high -end GPUs with sufficient dual -accuracy computing power. If the above -mentioned disconnection behavior is implemented, the scope of affairs will be relatively large.

In the past, the United States has "handled" three "hands -in" overhauls in China: In 2015, 4 Chinese institutions related to the "Tianhe II" project in China were included in the "entity list" by the United States; in 2019, Haiguang, Zhongke Shuguang, Wuxi Jiangnan, and Wuxi Jiangnan Five companies including the Institute of Computing Technology entered the entity list; in 2021, 7 super -algorithms including Feiteng, Shen Wei entered the entity list.

Zhu Jing said that from the above journey, it can be seen that for China's supercomputing, the means of American strikes to scope are facing upgrading. If the news is true, the method of blowing this time will be upgraded from "entering the entrance list of supercomputing related units" to "direct ban on products that can provide services for supercomputing services", resulting in the scope of affected from the field of supercomputing to the Internet field.

Zhu Jing said that the disconnection of high -end GPUs seems to be further blocked by China's supercomputing and intelligent computing, but the scope of affection is far magnified, and the disconnection of technical points must be considered to be involved in the upstream and downstream.

- END -