Xin Lang Cai Jing·2026-01-21 12:37Core Insights - The article discusses the contrasting energy challenges faced by the US and China in the context of AI development, highlighting that while the US is experiencing a severe electricity shortage, China has a significant surplus of power capacity [3][34]. - It emphasizes that the real bottleneck in AI development is shifting from GPU availability to the availability of gigawatt-level electricity and data centers [3][34]. Energy Supply and Demand - By 2030, the incremental electricity demand for AI development in China will only account for 1% to 5% of its new power generation capacity over the past five years, while in the US, it will account for 50% to 70% [35][4]. - In 2023, the US added approximately 51 GW of new power generation capacity, whereas China added an impressive 429 GW, showcasing an 8-fold difference in capacity expansion [37][6]. Efficiency and Cost Challenges - Despite having cheaper electricity, the energy cost for AI computation in China could be 40% higher than in the US due to efficiency issues in converting electricity into computational power [35][40]. - Chinese AI chips are limited to older manufacturing processes (7nm or older), while top US chips utilize advanced 4nm and 3nm technologies, leading to significant performance and efficiency gaps [42][41]. System-Level Solutions - The article suggests that while the US is attempting to innovate through decentralized energy solutions like small modular reactors (SMRs), China is leveraging its advanced ultra-high voltage (UHV) transmission technology to efficiently transport surplus renewable energy to AI centers [50][52]. - Chinese companies are integrating AI into energy systems, with examples like CATL embedding batteries in data centers for energy management, indicating a comprehensive approach to energy and AI infrastructure [52][51]. Global Competitive Landscape - The article posits that as global demand for AI increases, countries will face a choice between the US model, which requires significant investment in expensive chips and infrastructure, and the Chinese model, which offers a turnkey solution that includes both AI servers and the necessary energy infrastructure [53][56]. - The competition is framed not just in terms of chip technology but also in the ability to provide comprehensive energy solutions that support AI development [56][57].