AI能否带动电力提前跨越周期底部II:量化测算Token出海对中国电力的弹性
HTSC·2026-03-03 01:19

Investment Rating - The report maintains an "Overweight" rating for the public utility and power generation sectors [7]. Core Insights - The report highlights that the transition from the "training era" to the "inference era" in AI has significant implications for China's electricity demand, with potential elasticity exceeding 10% due to the global token consumption [2][5]. - It emphasizes the increasing importance of energy prices in the AI competition, suggesting that the cost of electricity will play a more critical role in the overall cost structure of AI models [3][5]. - The report recommends focusing on undervalued green electricity stocks and companies that will benefit from capacity price elasticity, particularly in the context of the anticipated slowdown in electricity supply growth starting in 2026 [1][6]. Summary by Sections Token Consumption and Electricity Demand - The report estimates that if the global daily token usage reaches trillions, the positive impact on China's electricity demand could be around 8% to 18% depending on the market share of domestic models [2]. - It notes that the elasticity of electricity demand due to token consumption is likely to be higher than that of electricity prices, particularly as the utilization rates of inference models are lower than those of training models [4][14]. Cost Structure and Electricity's Role - The analysis indicates that electricity costs currently account for about 5% to 10% of the total cost in AI data centers, with depreciation being the largest cost component [3][13]. - The report suggests that as the efficiency of domestic chips improves, the proportion of electricity costs in the total cost structure may continue to rise, potentially reaching 20% to 30% for self-developed chips [3][13]. Market Recommendations - The report recommends several stocks that are expected to benefit from the growth in renewable energy demand and capacity price elasticity, including companies like Longyuan Power, Huadian Power, and China Nuclear Power [6][8]. - It also highlights the potential for significant price increases in green certificates and capacity prices, which could benefit companies in the sector [4][6]. Market Dynamics and Competitive Landscape - The report points out that the market has not fully recognized the shift in AI competition dynamics, where the gap between domestic and foreign computing power is narrowing, and the demand for tokens is expected to grow exponentially [5][12]. - It emphasizes that while electricity prices are a factor, the core competitive advantage for domestic models lies in their cost-effectiveness and the ability to leverage local resources [5][12].

AI能否带动电力提前跨越周期底部II:量化测算Token出海对中国电力的弹性 - Reportify