How much electricity will U.S. data centers actually use by 2035?

According to BloombergNEF, data centers will require one-fifth of all U.S.-generated electricity by 2035, up from roughly 5% today. The absolute figure is projected at approximately 83 gigawatts of total power demand.

Why are forecasts being revised upward so frequently?

Multiple consulting firms have increased projections within months of each other because AI adoption is accelerating faster than models predicted. BloombergNEF's 2026 forecast is 83% higher than its own December 2025 estimate; EPRI more than doubled its 2024 figure; S&P increased its outlook by over 30% in six months.

Which U.S. regions will feel the strain first?

The PJM Interconnection (Virginia through Illinois) and ERCOT (Texas) are most vulnerable. PJM will allocate 34% of its electricity to data centers; ERCOT will dedicate 22% of generating capacity. PJM has already paused new grid connections twice and faces potential utility withdrawals.

What happens to electricity prices when grids become congested?

Scarcity drives up costs. PJM's capacity auction prices have climbed 76% year-over-year due to the supply-demand imbalance, even as new data center operators continue to compete for grid access.

U.S. Data Centers Will Demand One-Fifth of National Electricity by 2035, BloombergNEF Warns

The Electricity Crisis Facing AI Infrastructure

BloombergNEF projects that U.S. data centers will consume one-fifth of the nation’s electricity supply by 2035—a fourfold increase from current demand. According to TechCrunch AI, this expansion will require nearly 200 gigawatts of installed capacity, with roughly half allocated to large-scale model training and serving inference workloads. The concentration remains heavily weighted toward the United States: by 2033, the country is forecast to host 64% of all power-consuming AI chips globally.

The urgency of these projections reflects how quickly earlier estimates have become outdated. TechCrunch AI reports that BloombergNEF’s 2026 forecast climbed 83 percentage points higher than the same firm’s December 2025 prediction. The electrical industry nonprofit EPRI has similarly escalated its outlook—more than doubling its 2024 assessment—while Standard & Poor’s lifted its projection by over 30% between October and April. These rapid revisions underscore the pace at which chipmakers, hyperscalers, and AI startups are racing to secure compute resources.

Regional Grids at Breaking Point

The concentration of data center growth in specific interconnection zones is creating acute capacity crises. According to TechCrunch AI, the PJM Interconnection—spanning from Virginia westward to Illinois—will dedicate more than one-third of its total electricity output to data centers by 2035. ERCOT, the Texas grid operator, faces a similar predicament, with data centers consuming 22% of its generation capacity.

PJM’s situation is particularly strained. The organization temporarily suspended new connection requests for four years and only recently reopened its queue in April 2026. Despite this relief valve, one major utility operator, American Electric Power, has publicly signaled it may exit the interconnection entirely. The market response is telling: capacity auction prices in PJM have surged 76% year-over-year, yet data centers still represent 38% of all bids in the most recent auction—demonstrating that even elevated costs cannot deter the rush for AI compute infrastructure.

Global Energy Demand Follows U.S. Leadership

While North America remains the epicenter, AI’s electricity footprint will expand internationally. TechCrunch AI cites BloombergNEF’s projection that, under an aggressive AI adoption scenario, global data centers will inject 1,935 terawatt-hours of new annual electricity demand by 2033. That single figure approaches India’s total annual electricity consumption, illustrating the planetary scale of the infrastructure buildout.

Why This Matters

The energy crunch is no longer a long-term concern—it is reshaping capital allocation and grid policy today. Data center operators must now navigate interconnection queues measured in years, spot electricity prices approaching historic peaks, and the prospect of brownouts if generation capacity fails to keep pace with demand. Utilities face a binary choice: massively expand generation infrastructure (requiring billions in capital and regulatory approval) or restrict new data center connections, which would cede AI leadership to other regions. Policymakers in bottleneck zones like PJM must decide whether to fast-track renewable and nuclear generation projects or accept that U.S. dominance in AI compute may be constrained by physics and regulation rather than capital or talent.

For cloud providers and AI firms planning long-term workloads, site selection is shifting from labor costs and tax incentives to grid availability—a fundamental reordering of infrastructure economics that will echo through the 2030s.