Utility Load Forecasting When Data Center Demand Is Uncertain

Imagine a utility receives five data center requests.

Together they total 2 GW.

Should the utility add 2 GW to its load forecast?

Probably not.

Should it ignore the requests until construction starts?

That would be risky too.

This is the challenge facing utility load forecasting as data center development accelerates.

Large new loads do not behave like normal organic demand growth. One project can add hundreds of megawatts to an area that historically grew by a few percent each year.

At the same time, a request for service is not proof that the project will be built.

PJM is dealing with exactly this problem. In July 2026, its Board directed PJM to develop and maintain a Large Load Registry intended to improve load-forecast accuracy and transparency. PJM currently projects approximately 70 GW of new large-load demand by 2038.

The planning problem is therefore not simply forecasting higher electricity demand.

It is deciding which demand belongs in which planning case, when it should appear and where it will affect the system.

Start by separating requests from forecasts

A list of customer requests is useful information.

It is not a load forecast.

The distinction matters because projects can enter the pipeline at very different stages.

One developer may be conducting an initial site search.

Another may control land but have no customer commitment.

A third may have completed utility studies and begun construction.

Treating those projects equally would ignore information that is directly relevant to the likelihood and timing of load.

PJM’s existing forecasting methodology already considers project-specific information for large load adjustments and addresses the risk of double counting.

Utilities can apply the same basic discipline to their internal planning.

Build one version of the project pipeline

Large-load information has a habit of spreading across an organization.

Economic development has one list.

Transmission planning has another.

Distribution engineering has another.

Customer service may be tracking separate conversations.

Executives may have a list of announced projects.

Those records need to be reconciled.

For each large project, the utility should be able to see the requested capacity, site, proposed connection, target date, development status, load ramp and relevant commercial milestones.

The utility should also know who sits behind the project where that information is available and can appropriately be used internally.

Without that control, the same development can appear several times under different names.

Look for duplication before adding megawatts

Double counting can happen for perfectly legitimate reasons.

A hyperscaler may be comparing several locations.

A developer may submit alternatives through separate project companies.

A large campus may appear first as a full-build request and later as several phases.

Different utility teams may record the same customer independently.

If all of those entries are added together, the planning forecast can become detached from the underlying commercial reality.

This does not mean uncertain projects should disappear from planning.

It means the utility should understand how the projects relate to one another.

A project can remain visible in a high-growth scenario without being treated as firm demand in the base forecast.

Maturity should matter

Project size attracts attention, but maturity may be more important for forecasting.

A 700 MW inquiry with no defined development path should not necessarily carry greater planning weight than a 200 MW facility that has secured land, completed engineering and started construction.

Utilities can classify projects based on observable evidence.

At the early end, there may be little beyond an inquiry and a requested MW figure.

As the project develops, evidence becomes stronger: site control, formal studies, financial security, permits, customer commitments, equipment orders and eventually physical construction.

The exact framework will differ by utility.

The principle should not.

Planning treatment should become firmer as the evidence becomes firmer.

Timing needs the same scrutiny as probability

A project can be real and still arrive later than expected.

That matters for utility planning.

If 300 MW moves from 2028 to 2030, the ultimate demand may be unchanged, but the timing of transmission work, transformer purchases and generation requirements can change materially.

Large projects should therefore be tracked against development milestones, not merely the date originally entered on an application.

The forecast should also reflect the ramp.

A campus with an ultimate demand of 500 MW may start with 100 MW and expand over several years.

Planning the entire 500 MW into the first year can produce the wrong infrastructure timing.

Geography can overwhelm the system-wide forecast

A utility can also have an accurate total demand forecast and still plan the wrong infrastructure.

Location matters.

Two gigawatts distributed across a large territory is different from two gigawatts concentrated around three substations.

Large-load forecasting needs to connect demand to the electrical system.

Depending on the utility, that may mean grouping projects by:

• Substation
• Transmission zone
• Corridor
• Voltage level
• Municipality
• Planning area

This helps planners see where several individually manageable projects combine into one larger constraint.

It can also reveal where one infrastructure investment may support several customers.

One forecast is not enough

There is no reason to pretend that an uncertain project pipeline can be reduced to one perfect number.

Scenario planning is more useful.

A committed case might contain projects with strong evidence behind their schedules.

An expected case can incorporate a reasonable amount of less mature demand.

A high-growth case can test what happens if a greater share of the credible pipeline proceeds.

Utilities may also need a concentration case where several projects arrive in the same constrained area.

The purpose is not to produce artificial mathematical precision.

It is to understand whether today’s infrastructure decisions remain sensible under several plausible futures.

Tie capital decisions to evidence

The forecast becomes most useful when it tells the utility when to act.

A project may justify planning work long before it justifies construction.

A long-lead transformer may need an earlier decision than another piece of equipment.

Permitting may need to begin before every customer agreement is final.

Other capital can wait until the project reaches a stronger commitment milestone.

That means infrastructure planning should include explicit triggers.

For example, the utility might begin preliminary engineering at one maturity level but authorize major procurement only after additional customer commitments are in place.

This approach cannot remove stranded-cost risk.

It makes the reason for taking that risk much clearer.

Keep updating the evidence

Large-load pipelines move too quickly to review once a year.

Project schedules slip.

Customers choose another location.

Developers increase or decrease capacity.

Studies uncover constraints.

Construction begins.

Each event changes the forecast.

PJM’s planned Large Load Registry reflects the growing importance of maintaining better visibility into these projects rather than treating large-load information as a static annual input.

Utilities can apply that same discipline even when no regional registry requires it.

Better utility load forecasting does not require pretending to know the future

Data center development will remain uncertain.

The utility does not need perfect information to make better decisions.

It needs to separate a request from a commitment, identify duplicates, use realistic load ramps, understand where demand will appear and revisit assumptions as projects mature.

That produces a much more useful utility load forecast than simply adding every requested megawatt together.

PowerTek works with utilities and municipalities on load forecasting, transmission and distribution planning, large-load studies and long-term infrastructure planning.

The goal is not to predict exactly which data center will be built.

It is to make sound grid decisions even when the answer is still changing.