International Energy Agency
2026
2026
Official forecast: U.S. electricity use is expected to add more than 420 TWh through 2030, with data centers accounting for about half of the growth.
AI infrastructureTrackingReviewed 2026-07-09
AI Infrastructure: Deliverable Power
Official forecasts support rapid data-center electricity demand growth. The research question is whether power, interconnection, equipment, and commissioning can convert that demand into runnable capacity on the same timeline.
Evidence
Level 1/2 demand evidence; Level 3 transmission inference; Level 4 conditional capital scenario.
Scope
U.S. data-center electricity demand and the site-level delivery chain.
Boundary
This free ledger entry maps layers and evidence gates. Company-level candidates require separate current-data verification.
Signal map
Capital flow
Capital is being committed to AI compute, data centers, power procurement, generation, and grid infrastructure.
Core demand
Runnable compute capacity: capacity that is powered, cooled, connected, commissioned, and available on a defined timeline.
Delivery chain
Generation and procurement -> transmission and interconnection -> transformers and switchgear -> permits and construction -> testing and commissioning.
Decision variable
Deliverable power at the required site and date, not electricity or generation capacity in the abstract.
Capital relevance
Research-candidate exposures include electrical equipment, interconnection execution, site enablement, and time-to-energization. Specific companies and valuations require separate current-data verification.
Verification signals
- Official load forecasts continue to rise and are matched by signed, site-specific power arrangements.
- Interconnection milestones, permits, equipment delivery, construction, and commissioning advance on stated timelines.
- Critical electrical-equipment backlog, lead times, and pricing remain elevated relative to delivery schedules.
Disproof signals
- Official demand forecasts fall materially because of slower deployment or large efficiency gains.
- Interconnection, equipment, and commissioning timelines shorten enough for energized capacity to keep pace with demand.
Opportunity scenario: if runnable compute demand continues to grow while deliverable power and site enablement expand more slowly, layers controlling time-to-energization may face repricing pressure. This scenario should be withdrawn if the disproof conditions appear.
Primary and official sources
U.S. Department of Energy
2026
2026
Official reference: the LBNL 2025 update estimates a 2030 U.S. data-center electricity share of 11.8%, with a wide 9.5%-15.3% scenario range.
U.S. Department of Energy
Accessed 2026-07-09
Accessed 2026-07-09
Official operating context: large data-center loads are regionally concentrated, often require firm power, and can materially affect regional grids.