The Data Center Load Forecast Weapon: How Utilities Are Burying Ratepayers in Unbuilt Capacity

The Next Web recently highlighted how the power grid's interconnection backlog has swollen to more than 2,000 gigawatts of proposed capacity nationwide^[1], exceeding total US generating capacity. The framing is crisis: utilities need to build faster. But the real story is hidden in a prior clause: most of those projects will be withdrawn. That word, "most," is where the audit begins.

Here is how the data-center load forecast has become a rate-case weapon. A developer announces a gigawatt-scale facility. The utility models it into a 10-year demand projection. That forecast enters an Integrated Resource Plan or a capacity-expansion filing. State regulators, presented with "expected" load growth, approve capex for natural gas plants, transmission upgrades, and substation reinforcement. The utility then seeks cost recovery. Ratepayers fund the new assets. Years later, the facility is withdrawn, deferred, or moved to another utility's territory. The capacity sits idle or half-used. The cost remains on your bill, spread across the customer base, because it was never assigned to the customer class driving the spike.

The scale of speculative load is now measurable. More than $170 billion in AI data center capacity has been blocked, withdrawn, or stalled by community opposition since January 2024^[5]. That figure is not "blocked demand." It is capital that was shopped, speculated on, and never materialized. Yet the utilities that modeled it into load forecasts often kept the infrastructure approvals. Interrogate every data-center buildout story: Is the MW contracted (signed, collateralized, with a long minimum-take term)? Or is it announced, press-released, filed in an interconnection queue, and shopped to three utilities simultaneously, creating phantom load? The Electric Choice database tracks 3,300-plus data center facilities and 970-plus under construction or planned^[4]. How many of the "planned" have actual special contracts with binding ratchets and collateral? That number is redacted from most rate cases.

The Federal Energy Regulatory Commission has begun to notice. FERC found that PJM's data-center procurement plan may be unjust and unreasonable on cost allocation, transmission owner exit rules, and load-serving entity collateral requirements^[8]. Translation: the grid operator was on track to socialize the cost of data-center infrastructure across all ratepayers while letting the load itself exit cheaply or assign risk to transmission owners. FERC blocked it, but only partly. The deeper question remains unanswered in every docket: who bears the cost if the load forecast is wrong?

The solution is known and buildable now. Utilities can offer a curtailable or flexible interconnection service to data centers willing to accept brief grid-exceedance events, paired with bring-your-own-generation or storage mandates. The Department of Energy has published targeted actions across state policy, targeted actions, and grid planning mechanisms^[3] to address this mismatch. Regulators can demand a standing large-load tariff with cost isolation, high minimum-take ratchets (70 to 90 percent of contracted capacity over the asset life), collateral tied to unamortized investment, and transparent exit fees. Virginia, Ohio, and Oregon have filed versions; the protective elements are known. The window to intervene is now, in every state with a pending data-center rate case or tariff filing. Demand that the forecast be netted against historical realization rates, that phantom load be excluded, and that the customer class, not ratepayers, carry the risk of underperformance.