AI growth is testing whether households should remain protected when electricity supplies cannot meet every connected load.
Artificial intelligence expansion is rapidly increasing electricity demand. Reuters reported that very large customers could add about 70 gigawatts to PJM’s system by 2038. Data centers are expected to drive most of that growth across one of the nation’s most important regional power systems.
That demand creates a reliability question. When supply cannot meet every connected load, operators must reduce consumption or risk uncontrolled outages. The policy choice is whether flexible data centers should disconnect before homes, hospitals, and small businesses lose service.
PJM Interconnection is a regional transmission organization coordinating wholesale electricity across 13 states and Washington, D.C. It manages the high-voltage grid for 67 million people, balances supply and demand, and oversees regional markets. PJM does not own most generators or local distribution lines.
That distinction shapes its authority. PJM can direct generation, issue emergency instructions, activate demand response, and coordinate transmission owners. Local utilities and state regulators still control customer disconnections and service obligations.
Reuters reported that PJM plans to file 2 proposals with the Federal Energy Regulatory Commission. One would procure extra capacity after the latest auction missed its reliability requirement by 6.8 gigawatts. The other would create a registry showing where major users operate and how much electricity they consume.
The registry would improve information before emergencies. It would not automatically let PJM disconnect every facility. That limitation leads directly to the curtailment question.
Curtailment means temporarily reducing or stopping electricity use to protect the wider grid. A data center could shift to backup generation, reduce computing, or shut selected equipment. Advance agreements would define when curtailment begins, how long it lasts, and who carries the cost.
The proposal focuses on facilities that do not bring their own generation or secure equivalent supply. PJM’s connect-and-manage framework would let some new loads connect before enough generation exists if they accept interruption during shortages. Curtailment would occur before pre-emergency demand response.
The risk became concrete during July’s extreme heat. PJM Inside Lines reported an instantaneous load of about 162,700 megawatts on July 2, while roughly 6,000 megawatts of demand response suppressed the measured peak. PJM prepared utilities to curtail large loads with backup generation, although service remained uninterrupted.
PJM Heat Emergency Pushed Demand Near Record Levels. Created via Gemini.
That event showed curtailment is already possible. Formal agreements would replace improvised coordination with known obligations. The unresolved issue is whether earlier grid access should carry a weaker guarantee of uninterrupted service.
Connecting a data center can require substations, transmission upgrades, and additional generation. Those projects take years, while technology companies often seek power sooner. Accelerated connection can therefore transfer reliability pressure onto the existing system.
PJM’s latest capacity auction reached a price cap of $325 per megawatt-day but still failed to secure enough resources. Reuters reported that the shortfall increased blackout risk. Higher prices did not produce sufficient capacity on schedule.
PJM Capacity Shortfall Increased Regional Blackout Risk Exposure. Created via Gemini.
A reliability tradeoff could link faster connection with mandatory flexibility. Data centers receiving access before supporting supply arrives would reduce load first during emergencies. Facilities providing their own generation or firm capacity could receive stronger protection.
That arrangement would recognize that new connections carry different system risks. It could protect households from subsidizing reliability for exceptionally large users. Enforceable rules require cooperation among PJM, utilities, states, and federal regulators.
Households usually cannot shift essential electricity use to private generators. Small businesses may lose refrigeration, payment systems, communications, or production during outages. Rolling blackouts therefore impose costs beyond inconvenience. Households also receive little warning when emergency conditions force rapid operating decisions.
Data centers can provide essential services, including cloud computing, communications, finance, and government operations. Curtailment rules should distinguish critical functions from computing that can be delayed or moved. Agreements must verify that backup generation works without creating local risks.
Policymakers need standards for eligibility, notice, compensation, enforcement, and exemptions. State regulators must clarify whether utilities can interrupt specific large customers before rotating outages. PJM needs accurate load data and real-time coordination to decide safely.
The goal is not to punish artificial intelligence investment. It is to match grid privileges with reliability responsibilities before shortages become emergencies at scale. Clear rules can protect consumers while giving operators predictable choices.
The artificial intelligence buildout is reshaping America’s electricity system faster than planning can respond. PJM’s registry would make major loads more visible. Visibility alone will not decide who disconnects when supply falls short.
Curtailment agreements can place flexible industrial demand ahead of households during emergencies. They can reward facilities that bring dependable power or reduce consumption when called. Rules must protect critical digital services without granting every data center equal priority.
PJM’s proposal is a starting point rather than a complete solution. Utilities, states, federal regulators, and data center operators must define responsibilities before the next extreme-weather emergency. Faster grid access should not leave ordinary customers carrying the entire reliability risk.