OPINION: AI Pushed PJM to a Record. The Grid’s Safety Margin Is Shrinking

Introduction

On July 2, 2026, PJM Interconnection recorded peak demand of 166,147 MW, surpassing the 165,563 MW record from the 2006 heat wave, according to Pomegra. The system carried that peak. The warning is about the margin left around it. A record demand peak makes the grid’s spare room visible.

The July 2028–2029 capacity auction then showed a 6.8 GW reliability shortfall and $16.4 billion in capacity charges, with $6.3 billion attributed by Monitoring Analytics head Joseph Bowring to data-center demand.

The Record Arrived on July 2

PJM’s peak demand

On July 2, 2026, PJM reported by Pomegra reported 166,147 megawatts. The regional grid’s peak exceeded its prior all-time record. This dated entry concerns pjm’s peak demand, not a floating market slogan. Twenty years separated the two records; the new one is higher.

A peak-demand record puts pressure on planning because reliability must cover the hottest and most demanding moments. The practical consequence is a question of pjm’s peak demand, not a claim that every related measure moved in lockstep. The fact itself is a measured load, not a blackout. The distinction carries the argument.

The old benchmark

On 2006 heat wave, PJM reported by Pomegra reported 165,563 megawatts. That was the previous absolute PJM demand record. This dated entry concerns the old benchmark, not a floating market slogan.

The new peak exceeded it by 584 MW, making the difference concrete rather than rhetorical. The practical consequence is a question of the old benchmark, not a claim that every related measure moved in lockstep. The comparison does not state why every megawatt was consumed. The distinction carries the argument.

A Third Federal Grid Crisis

The crisis count

On 2026, Pomegra reported the third federally declared grid crisis. The event was described as the third federal grid crisis of the year. This dated entry concerns the crisis count, not a floating market slogan. AI data centers are a material driver, not a magical explanation for everything.

Repeated emergency conditions show a system under stress while stopping short of a claim that service everywhere failed. The practical consequence is a question of the crisis count, not a claim that every related measure moved in lockstep. The dossier links each crisis partly to growing AI data-center demand. The distinction carries the argument.

The partial AI link

On 2026, Pomegra reported data centers as a contributing demand source. The evidence says AI-center growth was part of the crisis context. This dated entry concerns the partial ai link, not a floating market slogan.

Partly matters: it allows a real responsibility to be named without pretending no weather or other load existed. The practical consequence is a question of the partial ai link, not a claim that every related measure moved in lockstep. No full allocation among causes is supplied. The distinction carries the argument.

Data Centers Are a Major Demand Driver

Their share

On 2026, industry forecasters reported about 50% of U.S. demand growth. The fact block says AI data centers now account for roughly half of total U.S. electricity-demand growth. This dated entry concerns their share, not a floating market slogan. Forecast demand is a warning light, not a meter already read.

That scale means data-center decisions can affect a public grid built for households and businesses alike. The practical consequence is a question of their share, not a claim that every related measure moved in lockstep. It is an approximate share, not an exact national meter reading. The distinction carries the argument.

The expected acceleration

On after 2026, industry forecasters reported strong further growth expected. Forecasters expect the data-center share to accelerate sharply. This dated entry concerns the expected acceleration, not a floating market slogan.

An expectation supports planning urgency, not a claim that the increase has already occurred. The practical consequence is a question of the expected acceleration, not a claim that every related measure moved in lockstep. The dossier does not set a precise future percentage. The distinction carries the argument.

EIA Projects Bigger National Use

The 2025 base

On 2025, EIA Short-Term Energy Outlook reported 4,195 billion kWh. U.S. electricity consumption reached that record level according to the EIA trajectory. This dated entry concerns the 2025 base, not a floating market slogan. Commercial load is changing the shape of the system.

The base shows that the coming forecasts begin from an already elevated demand system. The practical consequence is a question of the 2025 base, not a claim that every related measure moved in lockstep. The number is annual consumption, not PJM’s instantaneous peak. The distinction carries the argument.

The 2026 and 2027 outlook

On 2026–2027, EIA Short-Term Energy Outlook reported 4,269 and 4,399 billion kWh. The EIA projected those successive national totals. This dated entry concerns the 2026 and 2027 outlook, not a floating market slogan.

Forecasts make a capacity question visible, but they remain forecasts even when their direction is clear. The practical consequence is a question of the 2026 and 2027 outlook, not a claim that every related measure moved in lockstep. Neither projected value is a confirmed final annual reading. The distinction carries the argument.

Commercial Demand May Pass Residential

The projected crossover

On 2026, EIA Short-Term Energy Outlook reported commercial demand above residential demand. The dossier says this could occur for the first time in history because of data-center growth. This dated entry concerns the projected crossover, not a floating market slogan. A capacity auction turns future reliability into today’s price.

That reversal changes who sets the pace of electricity demand in the national system. The practical consequence is a question of the projected crossover, not a claim that every related measure moved in lockstep. The word should preserves the outlook’s uncertainty. The distinction carries the argument.

The driver named

On 2026, EIA Short-Term Energy Outlook reported data-center expansion. The record attributes the expected crossover to that expansion. This dated entry concerns the driver named, not a floating market slogan.

A grid designed around residential patterns must still serve homes when commercial load grows faster. The practical consequence is a question of the driver named, not a claim that every related measure moved in lockstep. The source does not quantify every customer class. The distinction carries the argument.

The Auction Put a Price on Reliability

Total capacity charges

On July 14, 2026, PJM reported $16.4 billion. The 2028–2029 capacity auction matched a record set late in 2025. This dated entry concerns total capacity charges, not a floating market slogan. Billions tied to data centers are not a theoretical cost.

Capacity charges are a forward reliability cost, making future scarcity a current bill issue. The practical consequence is a question of total capacity charges, not a claim that every related measure moved in lockstep. The reported total is auction-specific. The distinction carries the argument.

Data-center share

On July 14, 2026, Joseph Bowring of Monitoring Analytics reported $6.3 billion. The independent PJM market monitor attributed that portion directly to data-center demand. This dated entry concerns data-center share, not a floating market slogan.

A named share of the auction cost makes the distributional question impossible to dismiss as vague. The practical consequence is a question of data-center share, not a claim that every related measure moved in lockstep. The attribution belongs to Bowring, not a universal finding of every regulator. The distinction carries the argument.

The Cost Accumulates

Four-auction burden

On through the July 2026 auction, Monitoring Analytics reported nearly $30 billion. The dossier says data centers imposed that cumulative burden on PJM consumer rates across four auctions. This dated entry concerns four-auction burden, not a floating market slogan. A shortfall measures risk, not the hour of a blackout.

A multi-auction number reveals that the July price was not an isolated accounting event. The practical consequence is a question of four-auction burden, not a claim that every related measure moved in lockstep. The total is approximate. The distinction carries the argument.

The July capacity gap

On July 14, 2026, PJM reported 6.8 gigawatts. The auction fell short of what PJM needed to guarantee reliability at demand peaks. This dated entry concerns the july capacity gap, not a floating market slogan.

A deficit is a planning warning; it does not state that power failed on the auction date. The practical consequence is a question of the july capacity gap, not a claim that every related measure moved in lockstep. The use of “guarantee” is tied to reliability capacity. The distinction carries the argument.

Capacity Prices Changed Scale

The 2024 level

On 2024, Reuters reported $28.92 per megawatt-day. PJM capacity prices stood at that earlier level. This dated entry concerns the 2024 level, not a floating market slogan. A four-figure percentage increase changes the public bargain.

The before value makes the later escalation measurable rather than ornamental. The practical consequence is a question of the 2024 level, not a claim that every related measure moved in lockstep. It is not a retail residential price. The distinction carries the argument.

The current level

On 2026, Reuters reported $329.17 per megawatt-day. Reuters reported a 1,038% rise, mainly attributed to data-center growth. This dated entry concerns the current level, not a floating market slogan.

A jump of that size is a system signal, while “mainly” still allows other influences. The practical consequence is a question of the current level, not a claim that every related measure moved in lockstep. The report does not allocate every percentage point. The distinction carries the argument.

Factories Meet Regional Price Pressure

Pennsylvania industry

On December 2025 year over year, Reuters reported a 31% increase. Average industrial electricity prices rose by that amount in Pennsylvania. This dated entry concerns pennsylvania industry, not a floating market slogan. Industrial customers feel the regional gap.

Regional industrial users feel demand pressure through operating costs, not just policy debates. The practical consequence is a question of pennsylvania industry, not a claim that every related measure moved in lockstep. The period predates the July 2026 peak. The distinction carries the argument.

Ohio industry

On December 2025 year over year, Reuters reported a 26% increase. Ohio’s average industrial price rise exceeded the 7% national industrial increase. This dated entry concerns ohio industry, not a floating market slogan.

The difference shows that a national average can conceal a sharper local burden. The practical consequence is a question of ohio industry, not a claim that every related measure moved in lockstep. The facts do not calculate each factory’s response. The distinction carries the argument.

Households Are in the Same System

Virginia households

On the most recent year, EIA data reported by Fortune reported more than 13%. Residential electricity prices rose by that amount in the state with the most data centers. This dated entry concerns virginia households, not a floating market slogan. Residential bills also show the geography of data-center load.

The residential figure connects infrastructure demand to an everyday bill without claiming every increase has one cause. The practical consequence is a question of virginia households, not a claim that every related measure moved in lockstep. The source reports more than 13%, not an exact cents-per-kWh figure. The distinction carries the argument.

The public grid question

On 2026, the assigned fact block reported shared transmission and capacity. Households, factories, and data centers draw from a connected regional system. This dated entry concerns the public grid question, not a floating market slogan.

That shared setting is why allocation of new reliability costs becomes a public-policy issue. The practical consequence is a question of the public grid question, not a claim that every related measure moved in lockstep. The dossier gives figures, not a policy remedy. The distinction carries the argument.

Reliability Has Its Own Alarm

Exelon’s warning

On from 2027, Exelon CEO reported by Eastern Herald reported possible blackouts. The chief executive warned outages could occur as AI demand grows. This dated entry concerns exelon’s warning, not a floating market slogan. A level-three alert requires attention without predicting collapse.

A corporate warning identifies risk; it must not be rewritten as a promised national outage. The practical consequence is a question of exelon’s warning, not a claim that every related measure moved in lockstep. The fact block explicitly requires conditional treatment. The distinction carries the argument.

The earlier PJM shortfall

On December 2025 auction, PJM reported by Eastern Herald reported 6.5 gigawatts. That earlier gap preceded the 6.8 GW shortfall reported for July. This dated entry concerns the earlier pjm shortfall, not a floating market slogan.

Two deficits make the reliability concern more concrete without supplying a precise failure date. The practical consequence is a question of the earlier pjm shortfall, not a claim that every related measure moved in lockstep. A gap is not a blackout. The distinction carries the argument.

NERC Put 13 Regions on Notice

Level-three alert

On May 2026, North American Electric Reliability Corporation reported “Essential Actions”. NERC issued its most severe advisory category. This dated entry concerns level-three alert, not a floating market slogan. The 2027 warning stays conditional.

The severity calls for attention to reserves and planning, not panic language that outruns the advisory. The practical consequence is a question of level-three alert, not a claim that every related measure moved in lockstep. It is an alert framework, not an outage report. The distinction carries the argument.

The regional risk count

On May 2026, NERC reported 13 of 23 monitored regions. Those regions faced high risk of electricity shortages in summer peak conditions. This dated entry concerns the regional risk count, not a floating market slogan.

A high-risk assessment is a conditional statement about particular stress conditions. The practical consequence is a question of the regional risk count, not a claim that every related measure moved in lockstep. It does not say all 13 regions lost service. The distinction carries the argument.

The 2027 Claim Must Stay Conditional

PJM’s summer concern

On summer 2027, PJM context in Eastern Herald reported capacity may not be sufficient. The record says 2027 could be the first summer in which PJM cannot guarantee enough capacity for peak demand. This dated entry concerns pjm’s summer concern, not a floating market slogan. The grid still held. Its margin did not.

Could is the operative word: the risk is identified, the outcome is not confirmed. The practical consequence is a question of pjm’s summer concern, not a claim that every related measure moved in lockstep. No 2027 blackout is reported. The distinction carries the argument.

The correct public language

On 2027, the fact-block limitation reported a risk, not a certainty. The grid story becomes stronger when it tells readers what has been measured and what has only been warned. This dated entry concerns the correct public language, not a floating market slogan.

That discipline is essential where reliability anxiety can become rumor quickly. The practical consequence is a question of the correct public language, not a claim that every related measure moved in lockstep. The limitation is explicit. The distinction carries the argument.

Conclusion

PJM has not “gone down” in the assigned record. It has set a demand record, shown a capacity gap, and faced a high-level reliability warning while data centers drive a large share of demand growth. Those are serious facts without being a completed outage.

The 2027 blackout warning from Exelon’s CEO remains a warning. The public should not be asked to choose between complacency and panic. It should be shown the numbers that make planning urgent. The risk is named. The outage is not confirmed.

Sources

Primary sources

• U.S. EIA — Short-Term Energy Outlook — July 7, 2026

Secondary sources

• New York Times — PJM auction and data centers — July 14, 2026
• Reuters — EIA electricity-demand outlook — July 7, 2026
• Reuters — Data centers and power bills — July 7, 2026
• Fortune — Data centers and electricity costs — July 26, 2026
• Pomegra — PJM demand record — July 26, 2026
• Eastern Herald — Exelon 2027 warning — July 2, 2026