AI consumes 2.5% of US electricity

What AI 2027 Predicted

The scenario’s “KEY METRICS 2026” sidebar specifies that AI consumes 2.5% of US electricity by end of 2026, equating to approximately 33 GW out of 1.34 TW of total US generating capacity. This reflects the scenario’s depiction of massive AI datacenter buildout within the United States.

How We Track This

We monitor:

• EIA reports on total US electricity generation and consumption
• DOE estimates of datacenter and AI-specific electricity demand
• Utility and grid operator reports on datacenter interconnection requests
• Pew Research, LBNL, and other research organizations tracking datacenter energy footprint
• Distinction between total datacenter power and AI-specific share

Current Evidence

Total US electricity context:

• US total electricity consumption was approximately 4,082 billion kWh in 2024, with the EIA projecting 4,179 billion kWh in 2025 and 4,239 billion kWh in 2026 (EIA via Morgan Lewis)
• The EIA projects record power demand in 2026 and 2027, with AI and datacenters identified as the primary driver (Reuters, March 2026)
• US generating capacity is approximately 1.3 TW, consistent with the scenario’s 1.34 TW figure

Datacenter share of US power:

• Pew Research reports that US datacenters accounted for 4% of total US electricity in 2024, with demand expected to more than double by 2030 (Pew Research)
• The EIA’s Annual Energy Outlook 2025 estimates datacenters at 8% of commercial sector electricity in 2024, growing to 20% by 2050 (EIA)
• In Virginia, datacenters already consume 26% of the state’s electricity

AI vs. total datacenter:

• The key distinction: total US datacenter power was ~4% of national electricity in 2024. AI-specific power is a subset — probably 30-40% of datacenter workloads today
• 2.5% of US electricity ≈ 33 GW. Total US datacenter power was roughly 20-25 GW in 2024 (4% of ~500 GW average load)
• For AI alone to reach 33 GW (2.5% of US power) by end of 2026, AI-specific datacenter capacity would need to exceed total 2024 datacenter capacity — a very aggressive target

Growth trajectory:

• The DOE projects US datacenter energy demand doubling or tripling by 2028, suggesting AI’s share of national power will grow significantly but perhaps not to 2.5% by end of 2026

Sources:

• US power forecast — Reuters (March 2026)
• Pew Research: US datacenter energy
• EIA: Datacenter electricity trends
• DOE datacenter energy report

Counterevidence & Limitations

• Total datacenter power was ~4% of US electricity in 2024; AI-specific power reaching 2.5% (i.e., more than half of all datacenter power going to AI) is aggressive
• Grid interconnection and permitting create multi-year bottlenecks for new datacenter capacity
• Many announced datacenter projects won’t be operational by end of 2026
• Efficiency gains in AI inference (distillation, quantization, better hardware) partially offset demand growth
• The 33 GW figure appears to assume peak power, not average — average would be lower

What Would Change Our Assessment

• Upgrade to “on-track”: EIA or LBNL reports showing datacenter power exceeding 6-7% of US electricity by mid-2026 with majority AI workloads
• Downgrade to “behind”: Datacenter power growth stalls at 4-5% of US electricity through 2026
• Reinterpretation: If the metric includes all datacenter power serving AI-related cloud infrastructure (broader definition), the target is more achievable

Update History

Date	Update
2026-03	Total US datacenter power reached ~4% of national electricity in 2024. AI-specific share at 2.5% by end-2026 is aggressive but consistent with accelerating GPU deployment and new facility construction.