AI Data Centers Are Creating “Ghost” Power Demand on U.S. Grids

Proposed AI and cloud data centers are reshaping U.S. electricity planning—but a connection request is not the same thing as a data center that will be built and consume power. The gap between the two is increasingly described as “ghost demand”: prospective load that appears in utility and grid queues but may be duplicated, delayed or never materialize.

Reuters found that requests from very large electricity users, mostly data centers, exceeded 700 gigawatts (GW) across portions of the Midwest, Mid-Atlantic and South—more than 10 times industry estimates of current U.S. data-center power use. 1 That scale creates a forecasting problem with real consequences for grid reliability, infrastructure spending and customer bills.

What “ghost demand” means

A developer may seek capacity at more than one site, apply to multiple utilities, or reserve a place in a queue before it has secured financing, permits, equipment, land rights or a committed customer. Some proposals can also fail because transmission upgrades, water availability, construction schedules or on-site power plans do not work out.

Each request can be rational from an individual developer’s perspective: access to grid capacity is scarce and slow to obtain. But the aggregate queue can overstate how much electricity will actually be needed, where it will be needed and when it will arrive.

That does not mean AI and cloud demand is fictitious. It means raw queue totals are a poor substitute for a construction-ready load forecast.

The numbers driving the concern

The Reuters review put large-user requests in parts of three major U.S. regions above 700 GW. 1 In Texas, requests from data centers and other large users grew from roughly 48 GW in 2023 to more than 474 GW. 1

Those figures measure requested service, not electricity currently being delivered. Treating them as certain future demand could lead to a grid plan sized for facilities that never reach operation. Treating all of them as speculative, however, could leave a region unprepared if a meaningful share moves ahead quickly.

Why utilities are demanding proof of commitment

Utilities need a way to distinguish early interest from credible projects before they commit to substations, transmission lines and generation capacity. The most common tools make a reservation carry a financial or operational cost:

• upfront deposits or collateral;
• paid load or interconnection studies;
• minimum-bill and other contractual commitments;
• milestones tied to land control, permits and construction progress; and
• evidence of funding and a realistic energization schedule.

These measures do not prevent legitimate development. Instead, they give applicants an incentive to request capacity closer to what they can actually deliver.

Exelon provides a clear illustration. After filtering projects through transmission-security agreements that require financial commitments before major system investments, the utility’s large-load and data-center pipeline fell from 43 GW to 36 GW, a 16% reduction. 25

AEP Ohio’s data-center tariff likewise makes its process mandatory for new data-center service requests. Data centers and expansions of at least 25,000 kW must pay a load-study fee, and applicants must control the property through ownership, a lease or an option. 18

Why inaccurate forecasts affect everyone

Overbuilding can shift costs to customers

If a utility assumes every request will proceed, it may build excess wires, substations and generation capacity. Much of that infrastructure is long-lived, and its cost can be recovered through rates. Ohio’s consumer advocate has said an AEP-related settlement is intended to shield residential and small-business customers from unfair costs connected to infrastructure and transmission expansion for energy-intensive data centers. 15

Underbuilding can create reliability and price pressure

The opposite error also matters. If planners discount too many proposals and real facilities arrive on schedule, the grid may not have enough supply or network capacity. PJM, the grid operator serving 13 states and the District of Columbia, has said its widening supply-demand gap is being driven overwhelmingly by data-center demand and has advanced plans intended to secure additional power and manage the surge. 2

Capacity-market costs show why the quality of load forecasts matters. PJM’s independent market monitor estimated that existing and forecast data-center demand accounted for $29.4 billion in capacity-market charges across the operator’s four most recent auctions, according to reporting on the analysis. 29 That figure should not be read as a measure of ghost demand alone: it includes existing and forecast load, and the available evidence does not establish what portion stems from speculative requests. It does show that demand assumptions can flow through to customer bills.

Why states are auditing proposed projects

Texas and other jurisdictions are scrutinizing proposed data centers before scarce grid capacity and major upgrades are committed. The practical questions are straightforward:

• Who ultimately owns the project, and is financing credible?
• Does the developer control the land and have permits or a viable path to obtain them?
• Is there a realistic construction timeline and end customer?
• What power level, connection date and supply arrangement are being requested?
• What water use and local environmental constraints apply?
• If a site plans on-site generation, is it permitted, fuel-secure and capable of reducing its grid draw?

Pennsylvania’s experience highlights the difference between an announcement pipeline and projects nearing construction: Reuters reported that permits had been sought for only 20 of more than 100 proposed data centers. 1

The policy objective: a credible forecast, not a smaller queue

Data centers can be genuine large loads that require substantial new generation and transmission. The solution is not to assume every proposal is imaginary or to block development indiscriminately.

It is to convert a queue of tentative requests into a credible, time-specific forecast. Financial commitments, study fees, enforceable milestones and transparent project reviews help utilities reserve capacity for projects that can actually use it—and reduce the risk that households and smaller businesses pay for infrastructure built around demand that never arrives.