从中国返美，AI专家余震难消：这场竞赛或许已经结束

Core Insights - The article highlights the stark contrast between China's and the United States' capabilities in supporting AI development through energy infrastructure, with China being significantly ahead in terms of power supply reliability and capacity [1][2][4]. Group 1: AI Development and Energy Infrastructure - Data centers are crucial for the growth of the AI industry, with McKinsey predicting a need for $6.7 trillion in investments for new data centers globally from 2025 to 2030 to meet AI demands [2]. - In the U.S., the current state of the power grid is a major hindrance to the development of data centers, with Deloitte identifying grid pressure as the largest obstacle [2][5]. - In contrast, China has a robust energy infrastructure, with an average annual power demand increase that exceeds Germany's total annual consumption [2][4]. Group 2: Power Supply and Capacity - China's energy strategy has resulted in a power reserve margin of 80% to 100%, meaning its actual supply capacity is at least double the demand [4][5]. - The U.S. typically has a reserve margin of only 15%, which can drop even lower during extreme weather events, making it difficult to accommodate the additional load from AI infrastructure [5][6]. - China can utilize idle coal power plants to meet future AI energy demands, while the U.S. faces lengthy approval processes and local opposition for new power generation projects [5][6]. Group 3: Governance and Investment Models - The differences in energy infrastructure development stem from the contrasting governance models of China and the U.S., with China having a long-term, bureaucratic approach to energy planning [6][7]. - U.S. infrastructure projects rely heavily on private investment, which often seeks short-term returns, making it challenging to align with the long-term nature of energy projects [6][7]. - The pragmatic approach of Chinese policymakers allows for a focus on efficiency and results, while political divisions in the U.S. often stall infrastructure projects [7][8].