AI数据中心的快速负载波动正暴露出现有电力系统的深层缺陷。12026年7月22日，弗吉尼亚州阿什本的输电线路故障导致超过3吉瓦负载瞬间脱网，凸显了数十年未曾更新的数据中心电力堆栈在AI规模下的脆弱性。1AI数据中心在训练运行中可在毫秒内改变70%的负载，1而现有不间断电源（UPS）系统原本设计用于应对数分钟的停电，远无法应对全天24小时的快速波动。1
要解决这一难题需要从架构层面进行根本改革。1专家提出的方案包括三个步骤：将功率从低压上移至中压、从数据大厅转移到变电站附近的模块化机箱、以及将电池系统改为所有电流都必经的内联系统。1这种新架构已在美国能源部落基山国家实验室于2026年初进行了全规模系统测试，在真实电网故障和AI规模负载波动的同时影响下成功通过。1中压、模块化系统还能获得税收抵免，并在需量响应等电网项目中获得收益，为投资提供了经济激励。1
Rapid fluctuations in artificial intelligence data center power consumption pose a fundamental challenge to electrical grid infrastructure that cannot be solved by generation capacity alone.1 A transmission line failure in Ashburn, Virginia in July 2026 caused over 3 gigawatts of load to disconnect instantaneously from the grid, exposing critical weaknesses in decades-old data center power architecture operating at AI scale.1 A similar 2024 incident in Virginia demonstrated the vulnerability of current systems, when a single surge arrester malfunction triggered the simultaneous disconnection of approximately 60 facilities and 1,500 megawatts of load.1
The core issue stems from the speed at which AI systems alter power demand. During training operations, AI data centers can shift 70 percent of their load within milliseconds, a dynamic that existing uninterruptible power supply systems—designed to handle outages lasting minutes—cannot accommodate throughout a 24-hour cycle.1 To address these architectural deficiencies, researchers propose three key modifications: migrating power infrastructure from low to medium voltage, relocating modular cabinets from data halls to substations, and reconfiguring battery systems to operate as inline components through which all current must flow.1 Full-scale testing of this medium-voltage approach was successfully conducted in early 2026 at the Department of Energy's Los Alamos National Laboratory, where the new system performed reliably under simultaneous exposure to real grid faults and AI-scale load fluctuations.1 Beyond technical resilience, the modernized medium-voltage and modular architecture can qualify for tax credits and generate revenue through grid services such as demand response programs.1
评论
还没有评论，欢迎留下第一条。