CMSC: Collaborative driving of electricity consumption increases, public utility enterprises usher in value regression

CMSC released a research report stating that the overseas expansion of Tokens is driving the expansion of computing power, thereby leading to a significant increase in electricity demand. The bank estimates that by 2030, the proportion of electricity used by data centers in China's total electricity consumption is expected to reach around 4%. During the cyclical period, heavy asset, low obsolescence (HALO) public utility companies are experiencing a resurgence in value and are a typical HALO industry. In the environment of AI disruption and geopolitical turmoil, they are likely to become a certainty for investment. The main points of CMSC are as follows: - The rapid growth of AI computing power is leading to a significant increase in electricity demand for data centers. - According to data from the China Institute of Communications, from 2019 to 2024, China's data center electricity consumption is expected to increase from 824 billion kilowatt hours to 1,660 billion kilowatt hours, with an average annual compound growth rate of 15.0%. Although the current proportion of data center electricity consumption in the total social electricity consumption is low, reaching only 1.7% in 2024, the growth rate is significant. With more Chinese models going overseas, an increasing amount of overseas data is processed in domestic data centers. According to OpenRouter, the world's largest AI model API aggregation platform, the total Token consumption of the top ten models from February 16th to 22nd was about 8.7 trillion, with Chinese models accounting for 5.3 trillion, or 61%. Token overseas expansion represents the cross-border delivery of the value of electricity and is expected to further drive up data center electricity consumption. - By 2030, the proportion of electricity consumed by data centers is expected to reach around 4%. - Based on IDC's forecast, China's intelligent computing power (precision FP16) is expected to increase from 1,037.3 EFLOPS in 2025 to 2,781.9 EFLOPS in 2028, with an average annual growth rate of 38.9%. To study the electricity consumption growth brought about by the expansion of computing power, the bank has built a data center electricity consumption calculation model based on the electricity consumption of a single Token, considering the continuous decline in PUE. Assuming that China's intelligent computing power will have average annual growth rates of 25%, 35%, and 45%, by 2030, China's intelligent computing power is expected to reach 3,166 / 4,651 / 6,649 EFLOPS, and data center electricity consumption will reach 3,907 / 5,741 / 8,206 billion kilowatt hours, respectively. Assuming an average annual electricity consumption growth rate of 5.2% for the total social electricity consumption in China from 2026 to 2030, under different computing power growth rates, the proportion of data center electricity consumption in Chinas total social electricity consumption is expected to reach 2.9% / 4.3% / 6.1% by 2030. - The operation of data centers relies on stable and sufficient electricity supply, and the "dual carbon" target requires clean and low-carbon electricity. - In 2023, the National Development and Reform Commission issued the "Implementation Opinions on Further Implementing the "East Number West Calculation" Project to Accelerate the Construction of the National Integrated Computing Power Network", proposing that by the end of 2025, the proportion of green electricity in new data centers in national hub nodes exceeds 80%. Under the "dual carbon" target, data centers are expected to become an important force in promoting the absorption of renewable energy. In recent years, the green electricity trading market has become increasingly mature, and the premium for green electricity has gradually returned to a reasonable range. By locking in green electricity resources through long-term power purchase agreements, data centers can effectively avoid the risks of future fluctuations in fossil energy prices, achieving a win-win situation for computing power costs and green value. - During the cyclical period, public utility companies with heavy assets and low obsolescence (HALO) are experiencing a resurgence in value. - In the era of AI, there is constant technological disruption at the software level, but HALO assets are "visible and touchable" and have advantages such as high entry barriers, stable business models, predictable cash flows, and resistance to disruption by AI. Public utilities are a typical HALO industry and, in the environment of AI disruption and geopolitical turmoil, they are likely to be a certain choice for investment. Investment recommendation: Continue to favor traditional power leaders such as China Yangtze Power and GD Power Development, and pay attention to quality regional power targets such as Anhui Wenergy, Fujian Funeng, Shenergy, Guangxi Guiguan Electric Power, and Chongqing Fuling Electric Power Industrial. Risk warning: AI computing power demand is lower than expected, policy support is not as strong as expected, macroeconomic fluctuations.