Electric utilities have long planned the grid around a simple assumption: next year looks statistically like the last several, adjusted for weather and population. That assumption no longer holds. RMI's July 15 review of integrated resource plans found that 23 U.S. utilities updated their long-term filings in the first half of 2026, and the force behind nearly every material revision was large-load demand from data centers, a customer category that barely existed in the historical data most forecasting models were built on.
Nevada shows how far the shift has gone. Sierra Pacific Power and Nevada Power, which file a joint integrated resource plan as NV Energy, told regulators in an April 15 consumer session that the utility expects its electricity demand to double within four years and more than quadruple by 2045, with data centers accounting for over 80% of that growth. A model built on historical trend lines cannot absorb a single customer segment moving from a small share of load to the majority of it within two decades. NV Energy is not an outlier: RMI's dataset shows nine of the 23 utilities that filed IRPs in the first half of 2026 raised their load projections, four lowered them, and the group in aggregate pushed projected load up 3% in six months, the sharpest six-month swing since the initial wave of data center demand hit utility forecasts in 2024.
Why the Historical Forecasting Model Stopped Working
Historical load forecasting assumes demand changes slowly enough that a few years of data reveal a reliable trend. Large-load customers break that assumption, because a single data center campus can request more capacity in one interconnection filing than an entire utility territory added over the prior decade. Grid Strategies' National Load Growth Report attributes roughly 90 gigawatts of the 166 gigawatts of peak load growth currently forecast nationally to data centers specifically, a concentration in one customer type that older models were never built to isolate.
Few utility forecasts even separate data center load from other industrial demand today, which compounds the problem. Utilities are trying to size a customer category they cannot yet measure well, using tools designed for the slow, broad growth of homes and small businesses. That mismatch, more than any single bad forecast, is what pushed utilities to change methodology rather than simply update their numbers.
NV Energy's IRP Shows What Scenario-Based Planning Looks Like
NV Energy's April 15 filing walks regulators through a forecast built around named large-load scenarios instead of one trend line, and RMI cited it as the clearest example of the shift so far in 2026. Rather than asking how much demand will grow, the filing models low, medium, and high data center attachment cases and prices out the generation, transmission, and storage each one requires. That structure lets planners commit early to investments that make sense across every scenario while holding off on the ones that depend on which specific large-load customers actually sign contracts.
RMI's broader review found this scenario-based approach spreading well beyond Nevada. Most utilities that updated IRPs in the first half of 2026 built some version of scenario ranges into their filings, a departure from the single most-likely forecast that dominated planning as recently as 2023.
The Uncertainty Is Reshaping What Utilities Build, Not Just What They Predict
The forecasting shift shows up directly in construction plans. Utilities that updated IRPs in the first half of 2026 increased planned gas capacity by 9.4 gigawatts while cutting planned wind and solar additions by 17.3 gigawatts, changes several utilities linked in their filings to a need for capacity that can be built or deferred quickly as large-load demand does or does not materialize. Battery storage kept climbing regardless: planned battery additions reached 79 gigawatts by mid-2026, up from 66 gigawatts at the end of 2024, because batteries can be sited and scaled in modules that track uncertain load more closely than a large gas or nuclear plant can.
Several utilities, including NV Energy, Evergy Metro, and Dominion Energy South Carolina, also built advanced transmission technologies and virtual power plant programs (VPPAs) into their 2026 plans. Both add flexible capacity without locking in a twenty-year demand trajectory, and both let a utility respond to large-load requests competing for the same substation capacity without committing to a full-scale plant years before demand is confirmed.
What the Forecasting Shift Leaves Unresolved
Scenario-based planning solves the modeling problem, not the queue problem. California ISO's large-load issue paper, published January 30, found that transmission owners are already fielding more interconnection requests than their planning cycles were built to process, even after widening their forecast ranges. PJM Interconnection illustrated the opposite risk in January, when it lowered its near-term load forecast through 2032 after applying stricter vetting to speculative data center requests, a reminder that scenario ranges can move down as easily as up once utilities separate signed contracts from preliminary interest.
The open question for the rest of 2026 is whether utilities can tighten that vetting fast enough to keep pace with request volume, and whether state regulators approve the gas, battery, and transmission spending utilities say the uncertainty requires before the next IRP cycle locks in a new set of assumptions. FERC's ongoing review of PJM governance, with a technical conference scheduled for July 23, is one venue where that vetting and queue-processing question will be tested directly in the coming months.
Utilities are no longer asking how much electricity customers used yesterday. They are asking which proposed customers are credible enough to justify billions of dollars in generation, transmission, and substation investments. That shift may prove to be one of the most consequential changes to utility planning since electricity demand flattened two decades ago.
