A BloombergNEF report estimates that by 2035 data centers could use one-fifth of the electricity generated in the United States — roughly four times their current share.
AI-driven capacity growth
The consultancy predicts that surging demand for AI compute will raise data center capacity to nearly 200 gigawatts over the next decade. Close to half of that capacity will be dedicated to training and inference workloads, and much of these compute-intensive functions will remain concentrated in the United States. BloombergNEF forecasts that by 2033 the U.S. will account for 64% of AI chip power demand.
Estimates may be conservative
BloombergNEF’s new 2035 electricity-demand estimate is 83% higher than the firm’s December forecast. Other organizations have raised their projections as well: the Electric Power Research Institute (EPRI) has more than doubled its 2024 estimate, and S&P’s forecast increased by more than a third between October and April. These upward revisions reflect the rapid pace of data center development across the U.S.
Strain on regional grids
BloombergNEF expects most new data centers in the coming decade to connect to already strained power grids. The PJM Interconnection, which stretches from Virginia to Illinois, could see 34% of its electricity consumed by data centers, while ERCOT, covering most of Texas, may need to allocate 22% of its generating capacity to them.
PJM already hosts a large share of the country’s data centers and has struggled to handle connection requests from large generators and large loads. The grid paused applications for new sources to connect for four years, creating a precarious position as demand continued to rise.
Although PJM reopened its queue to new generating sources in April, the situation has become sufficiently strained that one utility, American Electric Power, has threatened to withdraw from the interconnection. The imbalance between supply and demand has pushed electricity prices up 76% over the past year.
Despite these congestion issues, data centers continue to pursue connections in PJM — they represented 38% of charges in the grid manager’s most recent capacity auction.
Global implications
Even though the U.S. is projected to host a majority of AI compute, data center growth will continue worldwide. BloombergNEF estimates that if AI adoption follows an aggressive path, data centers could create about 1,935 terawatt-hours of new electricity demand globally by 2033 — nearly the amount of electricity India uses in a year.
Why this matters
Rapidly rising data center demand poses significant challenges for power systems: it can strain grid capacity, raise electricity prices, and require new investments and regulatory responses, especially in regions where networks are already tight.
