Founder: The computing power needed for training large-scale models with trillions of parameters is in high demand. The distributed computing power domestically and internationally is about to see a sharp increase.

Founder releases research report stating that the gap between single-card memory and computing power is driving high demand for distributed computing. It is predicted that the proportion of distributed computing power to total computing power demand will increase from 15% in 2025 to 30% in 2030, with the proportion of distributed computing power increasing rapidly. The market space for distributed computing power is vast both domestically and overseas. Currently, containerized modular data centers are an important form of quickly deployable distributed computing facilities. It is recommended to focus on containerized modules, distributed power dispatch companies, and the synergy between computing and electricity. Founder's main points are as follows: Definition: Distributed computing power refers to the synergistic application of dispersed computing resources, which has better scalability, higher fault tolerance, and lower costs compared to centralized computing. Distributed computing power converts geographically dispersed computing resources into a logically unified computing pool through network coordination to collectively complete large-scale computing tasks. Containerized modular data centers are an important form of distributed computing facilities. Compared to centralized computing where computing resources are concentrated in one or a few core nodes, distributed computing power has the following advantages: 1) Better scalability: distributing computing tasks to multiple interconnected nodes can increase computing capacity by adding nodes to handle the workload. 2) Higher fault tolerance: the failure of individual nodes in distributed computing does not cause the entire system to fail, whereas in centralized computing, the failure of a core node affects the entire system. 3) Lower initial investment costs: centralized computing power requires purchasing high-performance core equipment and software, leading to higher initial construction costs, while distributed computing power can use ordinary hardware to form clusters, resulting in lower initial costs. Industry trends: The gap between single-card memory and computing power is driving high demand for distributed computing, and the proportion of distributed computing is expected to continue to increase. As model parameters grow from billions to trillions, memory usage will experience exponential growth, highlighting the gap between single-card memory and computing power. Distributed computing power has become the only solution as the physical limits of single nodes have been reached. Additionally, the construction cycle for distributed computing is shorter (18-24 months for centralized data centers compared to around 6 months for distributed), and it is more cost-effective (initial construction costs for distributed are 30%-50% of centralized), paving the way for a continuous increase in the proportion of distributed computing power. It is predicted that the proportion of distributed computing power to total computing power demand will increase from 15% in 2025 to 30% in 2030, with the proportion of distributed computing power rapidly rising. Market space: Both domestically and abroad are making efforts, with a vast market space for distributed computing power. 1) Domestically: China's computing power is being distributed from east to west, with intensive training tasks being placed in energy-rich western regions and real-time inference and low-latency tasks in eastern regions. Distributed computing has a faster response time, making it suitable for eastern regions, and its application space is expected to expand. 2) Japan and South Korea: Toyota, Honda, Japanese mobile games, VR rendering, and South Korean gaming nodes have high latency requirements, providing ample application space for distributed computing power. 3) Europe: The General Data Protection Regulation (GDPR) is the strictest data center protection law globally, aiming to provide high standards of protection for personal data within the EU economic area. Distributed computing power can distribute computing and data storage and processing, eliminating single point of failure and centralized attack risks, making it more secure and suitable for regulatory conditions in the European market. Containerized computing facilities: Integrated containerized power modules are the main application scenario. Currently, containerized modular data centers are an important form of quickly deployable distributed computing facilities. 1) Wetown Electric Group's Integrated Power Module (WMP): integrates core equipment such as medium and low voltage distribution, transformers, and UPS in standardized blocks or containers, improving space utilization by 40%, with a factory prefabrication rate of 85%, on-site installation reduced to 2 weeks, and significant energy savings. 2) CIMCPTU's CIMC Power Unit: through integrated design, it integrates power distribution, voltage conversion, power quality optimization, and safety protection functions to efficiently coordinate and distribute power from multiple sources such as mains power, UPS, energy storage, and diesel generators, supporting rapid power switching from multiple sources and saving approximately 60% of construction time compared to traditional solutions. Recommendations: 1) Containerized module advantages are significant, with great potential demand domestically and abroad. Recommended companies to focus on are: Wetown Electric Group (with MW-level containerized computing products and complete solutions), China International Marine Containers (received orders for modular data centers), Inspur Electronic Information Industry (prefabricated data center products), Kehua Data Co., Ltd. (integrated containerized solutions), etc. 2) Distributed power dispatch companies: Colorlight Cloud Tech, UCloud, Jiangsu Lettall Electronic, Jizhisuanli, etc. 3) Companies related to the synergy between computing and electricity: Far East Smarter Energy, Jiangsu Zeyu Intelligent Electric Power, Jiangsu Tongli Risheng Machinery Co., Ltd., State Power Rixin Technology, Zetta Group, etc. Risk warning: Technological advancements falling short of expectations, industry catalysis falling short of expectations, market competition risks, etc.