Disclaimer: The following content does not constitute any investment advice, guidance, or commitment, and is for academic discussion only.
Computing power chips are integrated circuits specifically designed to handle various computational tasks and provide core computational capabilities, serving as the cornerstone of the computing power industry.
Currently, domestic companies have gradually begun to use self-developed chips to train AI models, achieving good results in AI training and inference, with more computing power demands expected to be met by domestic chips in the future.
At the same time, the two leading domestic GPU companies, Moores Threads and Muxi Co., are about to enter the capital market, further promoting the development of the domestic GPU industry ecosystem.
With the iterative upgrades of AI large models and the continuous penetration of AI applications, the demand for computing power has surged, injecting strong growth momentum into the computing power chip market.
This issue provides an overview of the six major subfields related to domestic computing power chips and their associated companies, for further research reference.
Subfield One: GPU
The GPU, or Graphics Processing Unit, is a representative of high-performance computing parallel power chips, possessing overwhelming speed advantages in batch computing tasks, making it the “main force” in current AI training and inference calculations.
Associated Companies: Haiguang Information, Jingjia Micro, Loongson Technology, Chipone Technology, Dongxin Technology, East China Heavy Machinery, etc.
Subfield Two: CPU
The CPU, or Central Processing Unit, is akin to the “brain” of a computer, responsible for coordinating hardware resources, executing program instructions, and processing various data. It represents the general computing power chip field, adaptable to various computing scenarios, but its efficiency in large-scale parallel computing is lacking, resulting in weaker computing performance.
Associated Companies: Haiguang Information, Loongson Technology, Beijing Junzheng, Lanke Technology, China Great Wall, etc.
Subfield Three: ASIC
ASIC, or Application-Specific Integrated Circuit, is a custom chip designed for a specific application scenario or single function. Representative products in the computing field include AI chips and TPUs (Tensor Processing Units), sacrificing generality for extreme efficiency and energy efficiency, making it the “efficiency ceiling” of computing power chips.
Associated Companies: Cambricon, Yuntian Lifeng, Xindong Lianhe, Guokai Micro, Unisoc, Obsidian Light, Chiplink, Allwinner Technology, Runxin Technology, Chengdu Huamai, etc.
Subfield Four: Moores Threads
Moores Threads is a leading company in the domestic full-function GPU field, focusing on the independent research and development of full-function GPU chips and related products. It has successively launched GPU chips and corresponding boards such as “Sudi”, “Chunxiao”, and “Quyuan”, building a complete intelligent computing product line of “chip + graphics card + cluster”. The subscription was completed on November 24, and now it is just waiting for the listing and trading.
Associated Companies: Heertai, Zhongke Lanxun, Shenghong Technology, Maijie Technology, Jiechuang Intelligent, Donghua Software, Lihexing, Zhejiang University Network New, Haoshanghao, Zhuoyi Information, etc.
Subfield Five: Muxi Co.
Muxi Co. is a leading company in the domestic high-performance general GPU field, focusing on the independent research and development of full-stack high-performance GPU chips and computing platforms. It has successively launched the inference Xisi N series and the training-inference integrated Xiyun C series, while also researching the Xicai G series aimed at the graphics rendering market, covering multiple core scenarios such as AI computing and graphics rendering. The online subscription will start on December 5.
Associated Companies: Zhongke Lanxun, Chunzhi Technology, Guomai Culture, Zhongshan Public Utilities, Youfa Group, Chaoxun Communication, Kehua Data, Guanghuan New Network, Youke De, China Great Wall, etc.
Subfield Six: Huawei Ascend
The Ascend chip is Huawei’s self-developed dedicated NPU (Neural Processing Unit) for artificial intelligence scenarios. Currently, the main chips are Ascend 910 for AI training and Ascend 310 for AI inference. Huawei has also planned iterative products for the next three years, including 950PR, 950DT, and Ascend 960, 970.
Associated Companies: Tuowei Information, Changshan Beiming, Runhe Software, Ruantong Power, Shenzhen Huaqiang, Haohan Deep, Yuntian Lifeng, Duolun Technology, Aerospace Hongtu, Digital China, etc.
Follow me to get valuable information at the first time!
Disclaimer: The content of this article does not constitute any investment advice, guidance, or commitment, and is for knowledge sharing purposes only. Please read carefully. The market has risks, and investment should be cautious..