The Zhitong Finance app learned that Shouchuang Securities published a research report saying that demand for AI computing power has surged, and the importance of computing power scheduling and tuning has been highlighted. On the demand side, the large AI model has brought about a large amount of computing power requirements. At the same time, various emerging businesses such as application-side autonomous driving, face recognition, and intelligent manufacturing have put forward new requirements for flexible, convenient, and on-demand computing power. On the supply side, multiple policies support diversified computing power supply, and the “East Numbers West Computing” project has also put forward the need to achieve cross-regional scheduling of computing power resources. As computing power resources grow at an accelerated pace in the future, demand for computing power scheduling will also surge. At the same time, the complexity of large AI models and the high cost of computing power make it urgent to increase GPU utilization to release computing power, which in turn creates a demand for computing power tuning.

Investment advice:Targets related to computing power scheduling and tuning: Shenzhou Digital (000034.SZ), Zhongke Shuguang (603019.SH), Stoke (300608.SZ), Qingyun Technology (688316.SH), Zhongke Jincai (002657.SZ), Yuandao Communications (301139.SZ), Runjian Co., Ltd. (002929.SZ), etc.

The views of Capital Securities are as follows:

Computing power scheduling integrates multi-dimensional resources such as computing, storage, and networks at the bottom of the computing power infrastructure in the region according to the supply capacity of computing power resource providers and the dynamic resource requirements of the application demand side, and performs consistent management, integrated arrangement and unified scheduling of computing power resources based on the computing power scheduling platform, so as to achieve coordination and accurate matching of computing power resources across industries, regions, and levels. Computing power tuning improves GPU utilization by improving network communication capabilities and optimizing and upgrading software algorithms.

Computing power scheduling connects computing power leasing and model applications up and down. Operators closely collaborate with computing network resource providers to provide computing power support to the demand side. As the neural center of the computing power network, the computing network scheduling layer connects computing network resources and applications, connects the underlying computing resources downward, registers and signs them, analyzes the computing power requirements of terminal business scenarios upward, and intelligently breaks them down into various enabling platforms. Computing power scheduling lies between computing power leasing and model application, and is a key link running through hardware, leasing to models, and applications.

The state has issued a number of policies relating to computing power scheduling to support the strengthening of computing power scheduling capabilities. China attaches great importance to the development of the computing power industry. In order to optimize the supply of computing power resources and improve the overall level of computing power services, a number of policy documents on computing power scheduling have been issued. The “Three-Year Action Plan for the Development of New Data Centers (2021-2023)” clarifies the need to form a new data center development pattern with reasonable layout, advanced technology, green and low carbon, and compatible with the scale of computing power and digital economic growth. At the same time, it also points out the need to improve the supply of public computing power resources, optimize the computing power service system, and enhance computing power service scheduling capabilities. In February 2022, the country fully launched the “East Digital to West Computing” project to build eight national hubs for integrated computing power networks in Beijing-Tianjin-Hebei, Yangtze River Delta, Guangdong-Hong Kong-Macao Greater Bay Area, Chengyu, Inner Mongolia, Guizhou, Gansu, and Ningxia, while planning ten national data center clusters. Flexible scheduling of computing power between regions is required to open up computing power resources between east and west and east to achieve collaboration between east and west computing power.

Demand for AI computing power has surged, and the importance of computing power scheduling and tuning has been highlighted. Computing power scheduling is mainly responsible for matching the demand side of computing power with the supply side. On the demand side, the large AI model has brought about a large amount of computing power requirements. At the same time, various emerging businesses such as application-side autonomous driving, face recognition, and intelligent manufacturing have put forward new requirements for flexible, convenient, and on-demand computing power. On the supply side, multiple policies support diversified computing power supply, and the “East Numbers West Computing” project has also put forward the need to achieve cross-regional scheduling of computing power resources. As computing power resources grow at an accelerated pace in the future, demand for computing power scheduling will also surge. At the same time, the complexity of large AI models and the high cost of computing power make it urgent to increase GPU utilization to release computing power, which in turn creates a demand for computing power tuning.

Risk warning:Technological development falls short of expectations; macroeconomic risks; and the implementation process of AI applications falls short of expectations.