A new analysis from BloombergNEF warns that U.S. data centers will consume four times more electricity by 2035, accounting for one-fifth of the nation's total power generation. The surge is driven by an explosion in AI computing, which will push data center capacity to nearly 200 gigawatts over the next decade. Nearly half of that capacity will be dedicated to training and running AI models, with the majority staying concentrated in the United States. By 2033, the country is expected to host 64% of the world's AI chips by power demand.

These projections may actually be conservative. BloombergNEF’s latest estimate for 2035 electricity demand is 83% higher than what the firm predicted just last December. Other organizations have followed suit: EPRI, an electrical industry nonprofit, more than doubled its 2024 forecast, while S&P raised its projection by over a third between October and April. The revisions reflect the frantic pace of data center construction across the U.S., much of which is heading for already strained electrical grids. The PJM Interconnection, stretching from Virginia to Illinois, will see 34% of its electricity go to data centers, while Texas’s ERCOT will need to devote 22% of its generating capacity.

PJM has struggled to keep up with connection requests from both power generators and massive data center loads, even pausing new applications for four years. Though it reopened the queue in April, the situation has become so tight that utility American Electric Power has threatened to leave the interconnection. Electricity prices in the region have jumped 76% over the past year due to the supply-demand imbalance. Yet data centers remain eager to connect—they represented 38% of charges in PJM’s most recent capacity auction. Meanwhile, the global picture is just as stark: if AI adoption continues on its aggressive trajectory, data centers will create 1,935 terawatt-hours of new electricity demand worldwide by 2033, nearly matching India’s entire annual consumption.

Source: TechCrunch