According to reports, Nvidia, a U.S. chipmaker, will produce artificial intelligence (AI) chips designed specifically for the Chinese market, which is also known as the "special version for China". The chip was downsized from Nvidia's H100 chip to comply with the latest U.S. export control regulations.
In the past year, the breakthrough of artificial intelligence technology is the biggest highlight of the world's science and technology industry, and the development of artificial intelligence needs chip companies to provide computing power support. Some domestic manufacturers represented by Huawei have made significant progress in the field of artificial intelligence chips, and the gap between them and similar products of NVIDIA in terms of related hardware performance is not large.
However, there is still a considerable gap between the entire ecology of domestic artificial intelligence chips and foreign manufacturers. NVIDIA's AI chips are supported by a large number of third-party vendors, and NVIDIA provides a wealth of tools and libraries, such as CUDA (programming interface), CUDNN (deep neural network acceleration library), and TensorRT (software development kit for high-performance deep learning model inference), etc., which can help developers develop and deploy AI applications more efficiently.
NVIDIA has a large developer community and user base. Developers can access technical support, share experiences, and resources from the community to further grow the ecosystem. NVIDIA's AI chips are compatible with mainstream deep learning frameworks and development tools, such as TensorFlow, PyTorch, and more. This compatibility and interoperability makes it easier for developers to migrate existing models and applications to NVIDIA's platform.
For domestic manufacturers who need to purchase artificial intelligence chips, if they choose NVIDIA's architecture, they may not be able to choose the architecture of a domestic company, because it will mean huge upfront investment in talent reserves, technology compatibility, etc., and the migration cost cannot be ignored.
This time, Nvidia has engaged in the "special supply" of artificial intelligence chips to China, which also makes people feel a sense of déjà vu. In the past, when many domestic technologies made breakthroughs, foreign companies often relaxed some of their technological restrictions on China. Take, for example, EDA software in semiconductor design. In the early 90s of last century, 10 well-known universities in China, such as Tsinghua University, Peking University, Harbin Institute of Technology, Shanghai Jiao Tong University, etc., actively participated in the research and development of domestic EDA software, and successfully launched domestic EDA Panda software. After the successful development of the software, foreign countries suddenly lifted the blockade of China's EDA software, and implemented a series of strategies, such as donating software to some universities free of charge for teachers and students to use; Open up some of the tools to China and make them free; Establish a close alliance of interests with upstream foundries, and so on. These have led to the gradual marginalization of domestic EDA software in the domestic market.
Countless cases show that it is not that China cannot make semiconductor equipment, software and materials, but that the upstream field of semiconductors dominated by Western manufacturers has formed a relatively solid community of interests, and it is difficult for new leading companies to squeeze in. Taking lithography machines as an example, lithography machines require thousands of manufacturers to work together to tackle key problems. Because the price of lithography machine is extremely expensive, the supply exceeds demand, so it is not sensitive to the first machine, and lithography machine manufacturers generally rarely take the initiative to change the first business. At the same time, because the output of lithography machines is small, many lithography machine parts are relatively small in the global market, and a part may total a market of hundreds of millions of yuan, so even if the rising manufacturers develop the corresponding technology, it may not be cost-effective in terms of revenue.
In the past year, there has been good news in China's industrial chain, Huawei Mate60 series mobile phones and supporting Kirin 9000S series chips have made a major breakthrough in the case of full autonomy, and the supply exceeds demand. Huawei, as a major participant in the first chain, has made such a breakthrough, which has a significant driving effect on China's semiconductor industry. While developing mobile phone chips, Huawei also works closely with upstream manufacturers. Huawei not only provides them with stable orders, but also opens up its market knowledge and relevant data to upstream manufacturers of parts, equipment and materials. The binding of this kind of ** chain is likely to tear a hole in the existing community of interests in the West.
China is currently in a critical period of the development of domestic chips, such as semiconductors, new energy vehicles, industrial software, machine tools, aerospace, new materials and other fields. For domestic manufacturers, relatively speaking, the biggest difficulty in development is the breakthrough of the existing ecology and the construction of a new ecology. Overcoming this difficulty is not something that can be achieved by a single manufacturer, but requires the cooperation of relevant departments, enterprises and users, universities and research institutions to jointly give more opportunities and support to domestic manufacturers.
The good news is that many domestic technology companies, such as Alibaba and iFLYTEK, have purchased Huawei's Ascend series of autonomous and controllable intelligent computing chips. As long as the upstream and downstream of China's industrial chain work together, our ecosystem will soon develop.
China is the world's largest Internet member and manufacturing country, and is in the world's first echelon in the Internet industry, artificial intelligence, electronic manufacturing and other fields. We have every reason to believe that China's own industrial standards, which are based on China's local industry and user advantages, will also have considerable competitiveness in the world. (The author is a well-known observer of the technology industry and the author of "The Key to a Great Power").