US Capacity Additions Fall Short on Peak Coverage as PJM Demand Surges, ICF Says
ICF's new forecast shows 445 GW of US additions through 2030 contributing only 191 GW to peak load, as PJM demand is seen rising 43% by 2035.
ICF expects roughly 445 GW of new nameplate generation capacity to come online in the United States between this year and 2030, the consulting firm said in a report released Friday (2026-06-26). Because that buildout is dominated by solar and storage, which carry capacity factors well below conventional generation, only about 191 GW will actually contribute to meeting peak load — a 57% gap between what utilities will announce and what grid operators can actually count.3
The shortfall lands hardest in PJM, the grid operator covering thirteen states from Illinois to the Mid-Atlantic. ICF forecasts demand in PJM will rise 43% from 2026 levels by 2035. New York ISO is expected to see demand growth of just 14% in the same period, a divergence that underlines how unevenly the US power system is absorbing the data center buildout.3
Data centers are the primary driver. The IEA's global energy assessment found that in the US, data centers account for about half of all incremental demand growth. Global electricity demand from data centers rose 17% in 2025, with AI-specific consumption growing 50% in the same year, according to IEA data.1
PJM Western Hub spot power was trading at $42.83 as of Monday morning (2026-06-29), reflecting early-week demand conditions across the mid-Atlantic and Midwest.
Solar and storage, the two technologies dominating the additions queue, contribute to peak load coverage at a fraction of their rated capacity, since panels generate nothing at night and batteries are limited by discharge windows. ICF projects 68 GW of generation to come online this year, rising to 76 GW in 2027 and roughly 100 GW annually for each of the three years following, a buildout that nonetheless must run the gauntlet of a historically congested interconnection queue.3
Only 13% of the capacity that submitted interconnection requests between 2000 and 2019 had reached commercial operations by the end of 2024, according to IEA data. The other 77% had been withdrawn. Permitting backlogs, supply chain constraints, and queue management failures have made interconnection applications unreliable predictors of actual additions.1
The challenge plays out in real time. Elevate Infrastructure and ArcLight Capital Partners on Thursday (2026-06-11) brought online the 150-MW/600-MWh Prospect Power battery storage project in Rockingham County, Virginia, a PJM-region facility that ArcLight executives described as the product of more than two decades of investment in electric infrastructure. Battery storage is becoming an increasingly important component of grid reliability as demand accelerates across Virginia and the broader PJM footprint, company executives said.2
Whether that reliability contribution translates to sustained price support hinges on where demand actually lands. Signals on PJM real-time lean bullish, with data center load growth and the effective-capacity shortfall supporting the view. Yet if data center growth is front-loaded in developer announcements but slower to show up in metered load, near-term consumption could fall short of the projections embedded in the bullish case, leaving real-time prices below what forward markets are pricing for the structural story.1,3
ICF analysts note that supply and demand conditions vary considerably within regions, not just between them. The Southeast and New York regions could face capacity constraints similar to PJM's within a few years, the firm said, spreading the effective-capacity problem beyond the single grid currently absorbing the largest share of AI-driven load.3
If projects in the current queue follow the same trajectory as prior cohorts, where 77% of requests submitted between 2000 and 2019 withdrew before reaching commercial operations, effective additions would fall well short of ICF's 191 GW peak load contribution figure. With PJM absorbing the largest share of new data center load and forecast to grow at three times the pace of NYISO, the margin for error on interconnection throughput is tighter in the mid-Atlantic and Midwest than the national headline suggests.1,3