美国数据中心用电需求被上调36%，AI能耗飙升反而加剧泡沫担忧

Core Insights - BNEF predicts that by 2035, the power demand of data centers in the U.S. will surge to 106 GW, a 36% increase from previous estimates, driven by new early-stage data center projects [1] - The share of AI training and usage in data center capacity is expected to rise from approximately 12% to nearly 40% by 2035, significantly increasing data center utilization rates [1] - The aging and under-maintained U.S. power grid is projected to face a substantial power crisis due to the surge in electricity demand from data centers, new factories, and electric vehicles [1][2] Power Grid Challenges - Schneider Electric warns that maintaining current reserve capacity levels could lead to a significant power shortfall by 2028, with a potential gap of 175 GW by 2033, risking large-scale blackouts [1][2] - The North American Electric Reliability Corporation (NERC) indicates that data centers are already causing a notable increase in winter electricity demand, raising the risk of outages [2] Market Reactions - U.S. utility stocks have recently declined after reaching historical peaks, as investors grow concerned about the sustainability of returns amid rising electricity demand [4][5] - Companies like Constellation Energy and Vistra Energy have seen stock price drops due to slower-than-expected project announcements and adjustments to earnings forecasts [4][5] Investment Sentiment - There is a growing skepticism regarding the ability of utility companies to meet expansion promises, with concerns about potential capital waste on unfeasible projects [5] - The uncertainty surrounding AI as a profitable business model is causing traders to reassess investments in data center construction and operations, leading to discussions about an AI bubble [5][6] Technological Solutions - Some technologies and strategies are being implemented to alleviate grid congestion, such as the rapid increase in battery capacity in Texas [3] - Schneider Electric advocates for prioritizing grid enhancement technologies over large-scale new generation and transmission projects, which may not meet immediate power demands [3]