U.S. Data Center Power Demand Projected to Reach 106 GW by 2035
- AI Energy Demand Analysis Data Centers Grid Capacity
Marin Holt
- 0
- 7 minutes read
- BloombergNEF projects U.S. data center power demand will reach 106 GW by 2035, a 36% increase from its April 2025 forecast
- The U.S. had approximately 25 GW of operating data center power capacity in 2024
- Nearly one quarter of 150 new significant U.S. data center projects exceed 500 MW each
- PJM grid alone could see 31 GW of data center load by 2030, nearly matching all expected new regional generation
- AI training workloads now require 50-100 MW per facility; inference adds sustained baseload demand
U.S. data center power demand is on track to reach 106 GW by 2035, according to BloombergNEF projections released in early 2026. The forecast represents a dramatic acceleration driven primarily by artificial intelligence workloads, with hyperscale operators racing to secure power capacity through nuclear deals, renewable procurement, and behind-the-meter generation.
The Scale of the Problem
To put 106 GW in perspective, that exceeds the total installed generation capacity of most individual countries. AI training clusters alone now demand 50 to 100 MW per facility, with inference workloads adding sustained baseload demand that grows with each new model deployment. PG&E has mapped roughly [10 GW](/?p=4208) of new data center load requests across its service territory for the next decade.
The growth trajectory is steeper than any previous electricity demand driver. Unlike residential or industrial load growth, which historically increased 1-2% annually, data center demand doubles every 2-3 years. Interconnection queues across PJM, [ERCOT](/?p=4408), and CAISO now contain more data center capacity requests than all other load types combined in several regions.
How Hyperscalers Are Responding
The largest technology companies have moved aggressively to secure dedicated power supply. Microsoft signed a 20-year power purchase agreement with Constellation Energy to restart Three Mile Island Unit 1, an 835 MW nuclear plant in Pennsylvania. Amazon secured access to nearly 2 GW of nuclear capacity through multiple agreements. Meta signed nuclear power deals totaling up to 6.6 GW with Oklo, Vistra, and TerraPower as of January 2026.
Google and Westinghouse partnered to optimize the construction and operation of 10 planned nuclear reactors, with construction expected to begin in 2030. These commitments total over 10 GW of nuclear capacity dedicated primarily to data center operations.
The Grid Infrastructure Gap
Nuclear procurement alone does not solve the near-term problem. No small modular reactors operate commercially in the United States today. Permitting and construction cycles span five to seven years, placing first commercial units no earlier than 2030 with repeatable deployment in the 2030s.
This creates a critical infrastructure gap between 2026 and 2032 where data center demand growth outpaces new dedicated supply. During this window, operators rely on grid interconnection, battery storage, on-site natural gas generation, and demand flexibility programs to bridge the shortfall.
What This Means for the Industry
The 106 GW projection reshapes the entire energy infrastructure investment landscape. Transmission expansion, generation capacity additions, and distributed energy deployments all accelerate on timelines driven by data center commitments. Utilities that position to serve this load gain significant rate base growth. Those that fail to expand capacity fast enough risk losing economic development to competing jurisdictions.
For microgrid and distributed energy developers, data centers represent the highest-value customer segment. A single hyperscale campus requires more power than many small cities. Behind-the-meter solutions, including solar, storage, and on-site generation, provide both reliability and interconnection queue bypass, making them increasingly attractive to operators unwilling to wait years for grid capacity.
Sources: Utility Dive, Data Center Knowledge
Critical Analysis
A 106 GW national data center load driven by AI workloads represents the largest aggregate harmonic source in U.S. PJM, ERCOT, and CAISO interconnection queues now contain more data center requests than all other load types combined in several regions; 106 GW growth by 2035 outpaces generation and transmission expansion, and large power transformer lead times of 80-100 weeks create a compounding infrastructure bottleneck that cannot be resolved on the same timeline as data center construction schedules.
Critical Perspective
The 106 GW projection is built on current trajectory, not committed demand — a meaningful distinction when 40% of announced hyperscale projects have not yet executed power purchase agreements. Comparable market forecasts published in 2021–2022 for battery storage and EV charging overestimated actual 2024 deployment by 25–40% once interconnection queue delays and supply chain constraints were not modelled. The question energy professionals should be asking: what fraction of this forecast is backed by executed offtake agreements versus letters of intent?
