Electricity prices rose 11.5% in 2025 alone. Data centers drove 63% of PJM's capacity price surge. But cost isn't fate — it's policy. Understanding who decided this, how, and which proceedings are still open is how citizens change it.
Your electricity bill is not a fact of nature. It's the output of specific decisions made in specific regulatory proceedings — decisions that most ratepayers never knew were open for public comment. Understanding how those decisions get made is how you change them.
What's actually driving prices up:
PJM's Independent Market Monitor attributed 63% of the 2025/2026 capacity auction price increase — $9.3 billion — to data center load. Across three consecutive auctions: $21.3 billion in data center-related costs, roughly $315 per household across PJM's territory before state rate cases.
Natural gas prices rose 45% year-over-year as of October 2025. Because utilities can pass 100% of fuel costs directly to consumers through fuel adjustment charges, gas price spikes translate immediately to higher bills. The U.S. is exporting 25% more liquified natural gas than last year — increasing domestic costs.
The grid's aging infrastructure is being replaced — and ratepayers pay for it. Utilities requested more than $29 billion in rate increases in the first half of 2025 alone, double the amount requested in the first half of 2024. Much of this is legitimate infrastructure investment. Some of it is speculative — utilities building transmission for data center requests that may never materialize.
California, Texas, and Gulf states face infrastructure hardening costs from climate-driven events. These are real costs being legitimately recovered through rates — but they're often attributed incorrectly to clean energy or data centers in political arguments.
"Once they're up, we can't get electricity prices back down. Period."
— Energy economist at NYC Climate Week, 2025People are reasonably asking: why are we bearing these costs when the benefits of AI seem primarily to flow to tech companies and their shareholders? The answer is more nuanced than either side acknowledges. Some AI ROI is real and documented — drug discovery timelines cut from years to months, logistics optimization reducing fuel costs, grid management improving reliability. Some ROI claims are premature. And some of what's being built right now is speculative infrastructure ahead of returns that may not materialize. Citizens are entitled to ask which category a proposed facility falls into — and to demand evidence, not projections.
PJM Interconnection runs annual capacity auctions that determine how much generation will be available to meet peak demand three years ahead. The auction clearing price is what generators receive for being available, and those costs flow through to ratepayers. In 2024/2025, the clearing price was $28.92/MW-day. By 2026/2027 it had hit the FERC price cap at $329.17/MW-day — a 1,150% increase in two years. PJM's December 2025 capacity auction failed for the first time in history, falling 6,625 MW short of reliability targets. That failure is a grid reliability signal, not just a price signal.
The current system — where large industrial loads pay lower rates and their infrastructure costs are distributed to all ratepayers — was not designed with AI-scale data centers in mind. It was designed for a different era of industrial load, when major employers brought thousands of jobs and built communities around their presence.
The political economy was: large employers deserve rate incentives because they anchor local economies. The data center version of this bargain produces a different outcome: a 250,000-square-foot facility creates roughly 50 permanent jobs, receives rate incentives, and has its infrastructure costs socialized to the 40 million customer households that received rate increases in 2025.
Understanding this history matters because the arguments made to justify current policy — "these facilities create jobs," "they pay substantial taxes," "they need competitive rates to choose our state" — are the same arguments made for the industrial rate structures they replaced. Some of those arguments were true then. Some are true now. The question is whether the specific terms of the bargain serve the public.
Virginia's sales and use tax exemption for data centers, enacted in 2010, cost $1.6 billion in foregone state revenue in FY2025 alone — 79% of all state economic incentive spending. When the exemption was originally enacted, the projected cost was approximately $1.54 million per year. The gap between projection and reality is instructive: the incentive was designed for a different scale of development. The political will to revise it has not kept pace with the actual cost.
FERC's December 2025 ruling found that PJM's existing cost allocation is "unjust and unreasonable" — directing that costs must be allocated to those who cause them. This is landmark. It doesn't immediately change every state's rate structure, but it establishes the federal principle that data centers cannot continue to socialize their grid costs indefinitely.
Microsoft's January 2026 commitment to ask utilities to set rates high enough to cover the full electricity cost of its facilities — not passing those costs to residential customers — is the industry standard that should become the industry norm. One company doing this voluntarily matters less than all companies being required to.
AI is helping manage the grid — and creating the demand that strains it. Both are true simultaneously, and understanding the difference between them is essential to honest analysis.
What AI grid management is actually delivering: Google DeepMind's system achieves a consistent 30% reduction in data center cooling energy. Its partnership with PJM cuts interconnection approval timelines from years to months. Stanford spinout GridCARE freed up 80 MW of incremental capacity for Portland General Electric. A Yale analysis suggests AI grid optimization recovering 12–15% of currently wasted grid energy could theoretically offset AI's own demand growth.
The critical caveat: most of this optimization is proprietary — running on private facilities, benefiting private operators. Whether it gets embedded into the public grid infrastructure that serves all ratepayers is a policy question, not a technology question.
Demand response as a cost mechanism: Data centers can pause AI training jobs during peak demand periods, reducing grid stress. In Texas, ERCOT's demand-response contributions from large loads jumped from 2.7 GW to 13.3 GW for Summer 2026. This is real and valuable. It's also limited — hyperscalers sign firm power agreements specifically to guarantee they won't be curtailed on training workloads. The flexibility exists at the margins.
"Prices are rising across the U.S., and 2026 is very likely to bring further increases — up to 40% by 2030 compared to 2025."
— U.S. Energy Information AdministrationSome coal plant retirements are being delayed to meet data center demand — often at the administration's direction. Refurbishing a coal plant beyond its typical 40-year lifespan costs up to $1.3 billion, and coal operating costs increased 28% from 2021 to 2024. Those costs are passed to consumers. The irony: data center demand is creating pressure to keep the dirtiest, most expensive generation online, while delaying the cleaner alternatives. This is a cost impact and an environmental impact simultaneously.
Electricity pricing decisions happen in specific forums with specific timelines. Most citizens don't know these forums exist. That gap is part of how the current system persists.
The current cost structure is not a law of physics. It is a policy choice — made by legislators who approved tax exemptions, regulators who accepted cost socialization, and industry lobbyists who defended both. Naming it as a choice is the first step toward changing it.
The ethical standard is not complicated: those who cause costs should pay them. Those who use public resources should compensate the public for that use. Those who profit from infrastructure built with public investment should not have those costs shifted to the households least able to bear them.
The 21 million households behind on utility bills are not abstractions. They are families choosing between electricity and food, between keeping the lights on and paying for medicine. When utility shut-offs reached 4 million in 2025, some portion of that was directly attributable to rate increases driven by data center infrastructure costs that were socialized rather than assessed to those who caused them. That is an ethical failure with specific, traceable causes.
"Costs must be allocated to those who cause them and benefit from them."
— FERC Order, December 2025Virginia's new GS-5 rate class (effective January 2027) requires data centers above 25 MW to pay 85% of contracted infrastructure costs. Ohio's AEP established minimum take requirements. Georgia's PSC requires large customers to cover transmission construction costs. FERC's participant funding rulemaking has an April 2026 deadline. These are real changes, moving in the right direction. The pace is the question.
This section applies the AI Thinking Model™ — a framework for critical thinking, wisdom, innovation, strategy, and ethics developed by Liz B. Baker, Global Institute for AI & Humanity. Learn more →
