A new report from BloombergNEF projects that U.S. data centers will consume roughly one-fifth of the nation’s electricity by 2035 — about four times today’s level. The surge is driven by AI compute, which will push total data center capacity to nearly 200 gigawatts over the next decade. Nearly half of that capacity will go to training and inference workloads, and most will remain concentrated in the U.S. By 2033, the country is expected to host 64% of global AI chips by power demand.

These figures may actually be conservative. BloombergNEF‘s latest 2035 electricity demand estimate is 83% higher than what the consultancy projected just last December. Other organizations have also sharply raised their forecasts: EPRI more than doubled its 2024 estimate, and S&P’s forecast rose by more than a third between October and April.

The rapid buildout is straining already overtaxed electrical grids. BloombergNEF expects that most new data centers in the coming decade will connect to grids that are already under pressure. The PJM Interconnection, covering a swath from Virginia to Illinois, will see 34% of its electricity go to data centers. ERCOT, covering most of Texas, will need to devote 22% of its generating capacity. PJM, which already hosts a large share of U.S. data centers, paused new connection applications for four years and has struggled to handle requests from both large generators and large loads. The situation has become so severe that American Electric Power has threatened to pull out of the interconnection. Over the past year, the supply-demand imbalance has pushed electricity prices up 76%. Despite the congestion, data centers still represent 38% of charges in PJM’s most recent capacity auction.

Globally, if AI adoption follows an aggressive trajectory, data centers will create 1,935 terawatt-hours of new electricity demand by 2033 — nearly as much as India’s entire annual electricity use.