​According to PJM’s 2026 Long-Term Load Forecast, summer peak demand will rise from 160 GW in 2025 to 253 GW by 2046, a 58% increase driven primarily by data centers.

That growth has direct implications for battery revenues. Demand outpacing supply leads to elevated capacity payments, shown by PJM's last two capacity auctions. More frequent scarcity pricing events could also boost energy arbitrage revenues.

However, data centers draw consistent, round-the-clock power at a steady level so the spread between peak and off-peak demand is largely unchanged. Therefore, the arbitrage opportunity grows far less than it would from peak-concentrated demand additions like home electrification.

This article examines the methodology PJM uses to build its forecast, a key input in Modo Energy's production cost model and dispatch model for battery revenues across the Eastern Interconnect.

Without data centers, PJM peak demand would be shrinking. With them, it grows 35 GW in five years.

PJM's forecast combines a base model capturing residential, commercial, and industrial demand with a large load adjustment layered on top.

The large load component, principally data center additions, accounts for more than 100% of peak demand growth over the next five years as base demand contracts. PJM's large load adjustments grow by 35.1 GW between 2026 and 2031, against total demand growth of 34.6 GW.

Beyond 2031, the balance shifts slightly. The base layer begins to contribute and data center additions decelerate, but remain dominant. Summer peak demand reaches 253 GW by 2046, rising at 2.4% per year over the full period, with large load adjustments accounting for 78% of that growth.

Demand growth is concentrated: DOM, AEP, COMED, and PL all double load by 2046

Five zones spanning Virginia, West Virginia, Pennsylvania, Ohio, and Illinois at least double their annual energy demand by 2046: DOM (173%), PL (142%), AEP (124%), Dayton (121%), and COMED (100%). DOM, AEP, COMED, and PL have the largest data center pipelines, accounting for 74% of PJM's total annual demand growth between 2026 and 2046.

Notably, annual load rises much faster than summer peak demand. Data centers run at flat 24/7 utilization, lifting overnight and shoulder hours by a larger percent increase relative to the summer afternoon peak. The map above illustrates this directly: DOM shifts lighter when switching from the annual energy view to the summer peak view.

How PJM incorporates data centers: a rigorous filtering

Each July, PJM requests large load adjustment submissions from utilities. Utilities present requests to the Load Analysis Subcommittee (LAS) in September. PJM evaluates submissions through October and November, publishes a preliminary accepted list in November, and finalizes the forecast in January.

The ask-versus-accepted gap is significant. Raw utility submissions for 2030 totaled roughly 60 GW across the RTO. PJM accepted 34 GW, a 43% reduction before the forecast was published.

PJM applies four adjustments to every submission:

• Firm vs. non-firm classification: projects need an Electric Service Obligation (ESO) or Construction Commitment (CC) to qualify as firm unless otherwise specified. All other submissions are classified as non-firm.
• 70% utilization rate: applied to requested nameplate capacity as default, unless utilities provide historical data supporting a higher figure.
• Non-firm discounting: non-firm additions prior to 2030 are set to zero. Non-firm load from 2030 onwards receives a 50% haircut, with minor adjustments to match national average scaling.
• 36-month minimum ramp rate: demand from each project is phased in over a minimum of 36 months, imposed on all projects regardless of size.

Two zones received a different treatment than the standard 50% non-firm haircut:

• AEP submitted non-firm data already de-rated by more than 50% relative to its raw pipeline. PJM accepted those figures without applying an additional reduction.
• DOM's Dominion Energy submitted enough ESA or CC-backed projects to cover its entire modeled demand, classifying its portion as firm, though other utilities in the zone, such as REC, retain a non-firm component.

These are the two largest contributors to RTO-wide growth.

Firm load provides a partial floor, but uncertainty runs across both components

Firm additions are the most reliable part of the forecast. They reflect projects that have reached a later stage of development and are less likely to be cancelled. But those projects will all be built by the mid-2030s.

Non-firm additions join the forecast in 2030 and are less certain. The 50% haircut reflects PJM's own estimate of how many non-firm projects will be scrapped or delayed.

Later in the forecast, there are no specific large-load projects in any interconnection queue. PJM is extrapolating using national scaling benchmarked against third-party forecasts.

Both firm and non-firm projections carry downside risk. Projects that have already declared on-site generation have had their grid draw adjusted down accordingly. But as self-supply economics improve and potential policy reforms make behind-the-meter generation more attractive, more facilities may follow suit than the current pipeline suggests.

Connect and Manage (pending approval) adds a further layer of uncertainty. Projects subject to it would see their grid draw curtailed during stress events, shaving the summer peak without slowing the data center buildout more broadly. Neither Connect and Manage nor self-supply uptake are settled. Both could meaningfully pull realized demand below accepted figures.

The non-large-load components have a limited effect on demand shape

Beneath the data center layer, PJM's base model captures residential, commercial, and industrial demand using Moody's economic forecasts and Itron end-use data. Electric vehicle load, behind-the-meter solar, and behind-the-meter storage is sourced from S&P Global. Together these components are growing, but slowly relative to data centers.

Because data centers dominate the incremental load, the intraday demand shape is not changing materially. The curve shifts upward, but the relative pattern across hours remains broadly stable.

The limits of the forecast

The forecast is only as reliable as the projects underpinning it. Firm additions give the near-term outlook a solid foundation, but that foundation runs out in the mid-2030s. Beyond that, PJM is extrapolating and much can change between a data center pipeline and a data center drawing power.

Supply chain constraints and permitting delays can slow or prevent projects from reaching operation. Connect and Manage approval and self-supply uptake are the two live policy questions most likely to pull realized demand below accepted figures. Neither are settled, and both could meaningfully alter the capacity scarcity picture without slowing the data center buildout.