US data centers face massive power surge as AI demand accelerates

The surge in electricity requirements has forced analysts to repeatedly revise their forecasts upward. BloombergNEF’s latest estimate for 2035 is 83% higher than its prediction from just last December. Similar patterns have emerged across the industry: the Electric Power Research Institute has more than doubled its 2024 estimate, while S&P Global’s projections rose by over a third in just six months. This rapid escalation reflects an aggressive development pace that threatens to outstrip existing infrastructure.

Regional grids face the most immediate pressure. The PJM Interconnection, stretching from Virginia to Illinois, is projected to dedicate 34% of its electricity to data centers, while Texas’s ERCOT will likely commit 22% of its generating capacity. PJM has already struggled to manage connection requests, leading to a four-year pause on new applications. The imbalance has become so severe that American Electric Power has threatened to withdraw from the interconnection entirely. Despite these systemic bottlenecks, developers remain persistent; data centers accounted for 38% of the charges in PJM’s latest capacity auction. Globally, the trajectory is equally daunting, with AI-driven demand expected to generate 1,935 terawatt-hours of new consumption by 2033—an amount rivaling the annual electricity usage of India.