US Grid Capacity vs AI Demand
Regional analysis of the deliverable power gap between projected AI data center demand and committed grid capacity additions. National aggregates mask severe regional bottlenecks.
Last Updated: 2026-02-27
Data As Of: Dec 2025 (NERC LTRA) / Jan 2026 (PJM, ERCOT)
Sources: EIA AEO, NERC LTRA, PJM, ERCOT, MISO — all free, public, clickable.
US DC Demand 2028
37–66 GWavg
DOE/LBNL range
Net Capacity Adds/yr
~20 GW
EIA AEO 2026
Most Stressed Region
PJM
NERC assessment
Queue Wait (avg)
5.0 yr
LBNL 2025 ed.
Regional Grid Capacity vs Data Center Load
Data center load growth is concentrated in a few ISO/RTO regions. National capacity figures obscure the fact that the grid stress is highly localized.
| Region (ISO/RTO) | Current DC Load (GW) | Projected 2028 (GW) | Available Margin (GW) | NERC Risk |
|---|---|---|---|---|
| PJM | ~5.0 | ~12 | ~3 | elevated |
| ERCOT | ~3.0 | ~8 | ~5 | elevated |
| MISO | ~1.5 | ~4 | ~8 | moderate |
| Other | ~2.5 | ~6 | ~12 | low |
⚠️ Unit clarification
Regional figures use GW (peak capacity / peak load), not GWavg. Grid constraints bind at peak, not average. The national DOE/LBNL range (37–66 GWavg) is an average continuous draw metric and is not directly comparable to regional peak figures.
Deliverable Power Gap — National Summary
The "deliverable power gap" measures the difference between projected data center demand and the grid capacity that is realistically deliverable within the timeframe — accounting for interconnection queue delays, permitting timelines, and historical completion rates.
Gap Calculation:
Projected DC demand (midpoint): 52 GWavg (DOE/LBNL)
Committed capacity additions: 45 GW (EIA AEO 2026)
Deliverable gap: 7 GWavg (13.5% of midpoint demand)
Note: "Committed" means projects with signed interconnection agreements and financing. Queue backlog (~2,300 GW) is NOT committed — historical completion rate is ~14%.
Methodology — v0.1-beta
This page combines national demand projections (DOE/LBNL) with regional capacity data from ISO/RTOs and NERC reliability assessments to identify where the AI power gap is most acute.
Data Sources:
Known Limitations:
- Regional DC load projections are estimates — ISOs do not publish DC-specific load forecasts.
- GW (peak) and GWavg (average) are different metrics; direct comparison requires load factor assumptions.
- Behind-the-meter generation (on-site gas, solar) is not captured in ISO capacity data.
- This is an experimental research tool (v0.1-beta). Not investment advice.
📎 Cite This Data
AhaSignals. (2026). US Grid Capacity vs AI Demand. Retrieved from https://ahasignals.com/us-grid-capacity-vs-ai-demand/
Frequently Asked Questions
Can the US grid handle AI power demand? ▾
The answer depends on region and timeframe. DOE/LBNL projects US data center demand reaching 325–580 TWh by 2028 (37–66 GWavg). The US grid adds approximately 20 GW of net new capacity annually. The gap is most acute in PJM (Northern Virginia corridor) and ERCOT (Texas), where data center load growth is concentrated. National aggregates mask severe regional bottlenecks.
Which regions face the largest AI power gap? ▾
PJM (covering Northern Virginia, the world's largest data center market) and ERCOT (Texas) face the most acute pressure. NERC's 2025 Long-Term Reliability Assessment flags both regions for elevated reliability risk due to large load additions outpacing generation interconnection.
Is this investment advice? ▾
No. This page is an independent research audit produced by AhaSignals for educational purposes only. It does not constitute investment advice.
Related Pages
📊 Get Updates
Get weekly updates when the dashboard state changes materially, plus new research on consensus fragility and market divergence. Research-only. Not trade signals.
🔒 No spam. Unsubscribe anytime. 2,000+ researchers and practitioners as of Apr 2026.