华盛顿特区外一条电力线故障引发了对美国电网稳定性的严重担忧。[1]该事件导致数据中心在30秒内同时断开约3.1吉瓦负荷，这些设施代表当时PJM电网总需求的约3%。[1]电网随即出现电压波动，PJM系统额外电量峰值达3.49吉瓦，需要11分钟才能完全稳定。[1]
这起事件突显了大规模AI数据中心集中在北弗吉尼亚地区对电力基础设施造成的风险。[1]类似的并发故障在两年前曾经发生，2024年还出现过60个数据中心同时断开、拉低1.5吉瓦负荷的情况。[1]随着数据中心需求的急剧增长，这一问题日益紧迫。根据预测，到2040年，数据中心将占PJM电网总负荷的24%，相比目前的约6%增长四倍。[1]
为应对这一挑战，一些企业已在采取行动。ON.Energy正在四个数据中心园区安装总计3吉瓦的系统应对此类问题。[1]此外，电网运营商也在调整政策，德克萨斯州的ERCOT已要求大型负荷如数据中心必须具备"承载"电力中断的能力，以增强电网韧性。[1]
A downed power line outside Washington, D.C. triggered a cascade of disconnections across data centers in the region, with approximately 3.1 gigawatts of load shed simultaneously within 30 seconds [1]. The incident caused voltage fluctuations across the PJM electrical grid, resulting in visible light flickers throughout the area [1]. This event underscores a mounting vulnerability: the concentration of massive AI data centers in Northern Virginia poses an escalating risk to regional grid stability.
The disconnection represented roughly 3 percent of PJM's total demand at the time [1]. The grid required an additional 3.49 gigawatts of capacity to compensate for the sudden loss, taking 11 minutes to stabilize [1]. A similar incident occurred two years prior, and in 2024 alone, 60 data centers simultaneously disconnected and reduced load by 1.5 gigawatts [1]. Looking ahead, projections indicate data center demand will account for 24 percent of PJM's total load by 2040, compared to approximately 6 percent today [1].
To address this emerging challenge, ON.Energy is currently installing systems totaling 3 gigawatts of capacity across four data center parks [1]. Meanwhile, ERCOT has implemented requirements mandating that large-scale loads, including data centers, must "ride through" power interruptions rather than automatically disconnecting [1]. These coordinated efforts aim to reduce the grid destabilization risk posed by the rapid proliferation of energy-intensive artificial intelligence facilities in concentrated geographic areas.