Utility load forecasting has traditionally been an exercise in extrapolating slow-moving variables: population growth, economic output, weather, and the gradual uptake of electrical appliances. A single data center campus breaks that logic. It can add demand comparable to a mid-sized city over a few years, and it can appear in an interconnection queue long before local planners observe any shift in population or employment. That mismatch has turned data center load into a distinct forecasting problem rather than another input to an existing trend.
Forecasting data center demand requires estimating how much of a proposed load is likely to materialize and when. The scale of data center electricity demand is a separate question; here the focus is the forecasting methods utilities use to absorb it. A data center project differs from conventional load growth in size, timeline, and the probability that it moves from application to operation. This article explains how utility and system planners are adapting their methods, where the process still breaks down, and what that uncertainty means for grid investment decisions.
The Limits of Traditional Load Forecasting
Most utility load forecasts are built around econometric and end-use models. Econometric approaches relate historical demand to variables such as GDP, population, and employment. End-use models aggregate the expected consumption of buildings, industrial processes, and appliances. Both methods assume that future demand follows patterns similar to the past, with weather normalization removing temperature-driven variation to reveal underlying growth.
Data centers do not fit these assumptions. Their demand is not primarily weather-driven, although cooling systems respond to outdoor temperature. It does not scale with local population or economic output in the usual way. A hyperscale facility serving global cloud workloads can locate in a rural area with little local economic change and then connect a large block of load in a single step. Traditional models are built around gradual, continuous growth, so step changes of this kind fall outside their design.
The IEA has noted that data center electricity demand is becoming a more visible component of load growth in several markets, even where total demand has been flat or declining. That visibility is putting pressure on forecasters to move beyond trend extrapolation. Where data center development is concentrated, the interconnection queue itself has become a source of forward-looking information, though not a simple one.
The timing problem matters just as much. A data center project can move from initial application to full load in a couple of years or remain in an interconnection queue for much longer. Some developers file multiple applications for the same prospective campus to preserve optionality. A planner cannot simply add queued capacity to a forecast; each application carries a different probability of becoming real load. That probability depends on land control, financing, power purchase agreements, and other information utilities often do not receive.
From Request to Operating Load: The Project Stages That Matter
Data center demand moves through several distinct stages, and a useful forecast tracks the probability of each transition rather than treating the initial request as final. Utility planners commonly distinguish six milestones:
- Requested load: an interconnection application or site inquiry; high uncertainty, often including multiple speculative requests for the same campus.
- Contracted load: a signed power supply or interconnection agreement; a stronger signal but still not certain to materialize.
- Construction: physical work has begun; schedule risk remains from equipment and labour constraints.
- Energized load: the facility is connected and begins drawing power, typically at a partial load during commissioning.
- Ramped load: IT equipment is installed in phases, and demand increases stepwise over months or years.
- Full operating load: the facility reaches its designed capacity, which may occur long after energization.
A request for several hundred megawatts and a forecast of several hundred megawatts are not interchangeable. Utility planners often assign probabilities to each milestone and multiply those probabilities by the load associated with the stage. A project with a signed contract but no construction start may enter the medium-term forecast at a fraction of its eventual size; a project already ramping enters near-term operational plans at close to its observed demand.
Even after the likely project load is established, forecasters separate energy consumption from peak demand. A data center’s contribution to system peak is determined by its load at the time of the system peak, not by its maximum draw. Coincidence factors vary by region and by the shape of surrounding load. A facility with a high, flat load profile may contribute nearly its full demand at system peak, while one with strong diurnal variation may contribute less. That distinction matters because transmission and capacity decisions depend heavily on peak conditions.
Three Forecast Horizons
Planning horizons add another layer to that distinction. A horizon is set by the question a study is designed to answer, not by a fixed category: transmission expansion studies often reach decades ahead, distribution and resource planning operate on shorter cycles, and operational scheduling sits closest to real time. The level of project certainty each step can assume changes along that spectrum.
Long-Term Transmission Studies
Ten- to twenty-year transmission planning relies on broad scenarios rather than project-level certainty. Data center growth appears as a scenario input, not as a specific list of queued projects. A high scenario may assume most prospective campus capacity connects; a low scenario may assume only contracted projects proceed. These alternatives feed transmission expansion and resource adequacy studies, where differences in data center growth can change the need for new capacity by a material margin.
Medium-Term Forecasts
In the two- to five-year horizon, planners assign probabilities to projects by milestone. Executed contracts, construction status, and financing arrangements become direct inputs. As a project passes through the stages described above, its weight in the forecast increases. This is where the distinction between requested and contracted load becomes most important, because medium-term forecasts often inform distribution substation upgrades and regional transmission reinforcement.
Near-Term Operational Planning
Months to two years ahead, operational planning depends on executed agreements and commissioning schedules. Dispatch, maintenance, and outage coordination depend most directly on projects that are energized or near-energized, though earlier-stage projects can also factor in when their executed agreements, schedules, or expected commissioning dates carry near-term planning consequences. Behind-the-meter generation complicates this picture. A data center with on-site generation may not appear in the utility’s retail sales forecast, but the utility still needs to plan for transmission service, backup supply, and the possibility that the on-site plant trips. The net load the system sees can be far lower than the data center’s actual consumption, while reserve and contingency requirements remain.
Where the Process Breaks Down
Queue Data Quality
Applications can be duplicated, withdrawn, resized, or transferred between entities. In the United States, FERC’s generator interconnection queue reforms under Order No. 2023 are intended to improve public reporting for generation projects. For FERC-jurisdictional transmission providers, large-load requests are generally handled through load-interconnection procedures, with the specific sequence set by the applicable tariff. For large loads, however, the equivalent project-level visibility often has to be developed through utility-specific processes.
A request may enter the queue at a size far larger than what is ultimately built, only to be scaled back into smaller phases or withdrawn entirely. If a utility treats queue volume as a point estimate, it can overstate near-term demand. If it discounts queue data too heavily, it can miss a project that is already purchasing transformers and switchgear.
Timeline and Opacity
A data center developer may secure a grid connection but then delay construction because of supply chain constraints, financing, or customer demand. Meanwhile, the utility must decide when to include the load in its planning horizon. Including it too early can lead to premature infrastructure investment; including it too late can leave the system short of capacity. Part of the problem is coordination between two industries with different planning cultures.
Opacity compounds the timeline problem. Data center developers often treat IT load, server configuration, and cooling design as commercially sensitive. Utilities may know the contracted capacity but not the expected load factor, ramp rate, or schedule for adding phases. This forces planners to rely on generic assumptions that may not match the facility’s actual operation. The result is a forecast that is directionally correct but potentially wrong on the margin where investment decisions are made.
Geographic Resolution
Local transmission constraints can be hidden by aggregate forecasts. A state or regional forecast may show modest load growth, while one county hosts several large campuses and strains a specific substation. On their own, system-wide forecasts may not contain enough geographic detail to support distribution and subtransmission planning. Without geographically detailed load forecasts, utilities can miss the exact location where a transformer or line upgrade is needed. This is part of why electricity availability and grid constraints shape location decisions.
What Comes Next
Utilities and developers are moving toward more formal data-sharing arrangements. Some jurisdictions are discussing requirements for standardized project status updates, including expected in-service dates, phase sizes, and load profiles. These discussions reflect a recognition that forecasting cannot improve without better information from the entities that control the load.
Probabilistic methods are also spreading. Instead of treating a queued project as either in or out, planners assign probabilities to different stages and update them as milestones are reached. This approach produces a distribution of possible future loads rather than a single line. It fits naturally with scenario-based transmission planning and resource adequacy assessments, where the goal is to understand risk rather than predict a single outcome.
Integration across planning functions is another area of change. Historically, distribution planning, transmission planning, and resource adequacy have used separate load forecasts. Data centers make that separation harder to sustain, because the same project can affect all three at once. Utilities are moving toward more integrated forecasting, where a single set of assumptions about data center development feeds every planning process. This can reduce inconsistencies, though it also requires organizational changes that are not always quick.
Regional variation persists. In some Asian markets, grid operators and utilities are forecasting data center growth alongside wider electrification and industrial expansion. In North America and Europe, the challenge is shaped by queue backlogs, market structures, and the availability of historical data. European transmission system operators, through ENTSO-E’s scenario processes, are beginning to treat data center demand as a distinct input rather than a residual in long-term planning.
Planners who model large loads as contingent project portfolios, with probabilities that change as milestones pass, can communicate residual uncertainty more honestly. The practical aim is to make that uncertainty explicit enough to guide investment decisions. Treating data center demand as a dynamic, project-level input rather than a macroeconomic trend gives utilities a more useful basis for planning than reliance on historical patterns alone.
References
- IEA — Electricity 2025: data centre electricity demand as an emerging driver of power system planning.
- FERC — Order No. 2023: generator interconnection queue transparency and the separation of large-load interconnection.
- ENTSO-E — Ten-Year Network Development Plan scenario framework: data centre demand treated as a distinct long-term planning input.
