AI数据中心压垮美国最大电网，13州6700万人或遭轮流停电

Core Insights - The surge in AI demand is driving a severe supply crisis for PJM Interconnection, the largest nonprofit grid operator in the U.S., which serves 67 million people across 13 states [1] - PJM is facing a supply-demand imbalance, with electricity demand expected to grow at an annual rate of 4.8% over the next decade, a significant challenge for a system that has seen stagnant growth for years [1][2] - The closure of aging power plants is outpacing the construction of new facilities, increasing the risk of rolling blackouts during extreme weather events [1][2] Supply and Demand Dynamics - The demand for electricity in Northern Virginia has reached unprecedented levels due to the influx of data centers, with Dominion Energy reporting requests exceeding 40 gigawatts, enough to power at least 10 million homes [2] - By 2039, peak demand in the Virginia network is expected to double, while supply is shrinking due to environmental policies leading to the closure of coal and gas units [2] Political and Management Challenges - PJM faces political challenges as different states have varying policies on generation types, construction permits, and price regulation [3] - Pennsylvania's Governor has filed complaints to control rising prices, while other state governors threaten to withdraw from PJM due to dissatisfaction with its market mechanisms [3] - The departure of PJM's long-time CEO adds to the uncertainty, with the interim CEO acknowledging the need for coordinated policies and investments to meet demand growth [3] Regulatory and Industry Tensions - PJM proposed a plan to cut power to data centers during extreme grid stress, which faced strong opposition from tech giants like Amazon and Google, who argue it is discriminatory [4] - The inability to reach a consensus among tech companies, power suppliers, and regulators has stalled the development of related rules [4] - An independent monitor has warned that unless data centers can generate their own power, PJM will face challenges in reliability and may need to allocate blackouts [5]