AI data centres bring eminent domain into the power debate

On July 16, 2026, a legal analysis brought a less visible consequence of AI expansion into focus: transmission lines needed by data centres can cross private land, and US utilities may be able to use eminent domain when voluntary agreements fail. That is not an automatic authorisation to take property, nor is it a power created specifically for AI. It is a debate about applying longstanding legal powers to a new source of electricity demand.

What eminent domain means

Eminent domain allows a government to acquire private property for public use, with compensation, even without the owner's consent. In many places, that authority can be delegated to electric utilities to build grid infrastructure. But delegation does not settle every case. A company still has to follow applicable procedures, while public purpose, compensation and state-law requirements can all be reviewed.

That is why two claims need to be kept apart. A transmission line serving an area with new data centres may require easements or land. It does not follow that every line serving a private data centre automatically meets the public-use test. The July 16 analysis asks exactly that question and stresses that state rules and the details of a project matter.

AI is becoming a new kind of grid load

The issue has become sharper because AI data centres are not just another electricity customer. The US Department of Energy describes large training facilities as sites where thousands of chips operate in coordinated cycles, creating a dynamic load that the grid must monitor and absorb. In a transmission-needs study released for public consultation in July 2026, the department lists the growth of hyperscale AI data centres among the factors driving the need to expand and modernise the grid.

That study neither chooses a route nor authorises a particular taking. It does provide the context: after decades of relatively slow electricity-demand growth, planners now have to account for large industrial loads, electrification and data centres at once. Transmission lines, substations and generation are becoming a material part of the competition for computing capacity.

Federal land, on-site power and the wider grid

On July 20, the US National Nuclear Security Administration announced that it had selected Amentum to negotiate development of a one-gigawatt AI data centre and on-site energy source at Savannah River in South Carolina. The proposal includes about two gigawatts of generation, beginning with gas and moving toward nuclear power. The agency also made clear that this is not a final lease award: negotiations, permits, safety reviews and other approvals remain.

That project uses federal land and pairs power with the data centre, so it does not show how a dispute with private landowners should be decided. It does illustrate why developers are trying to combine land, generation and computing. Producing more electricity at a site may change pressure on some lines, but it does not erase grid needs or local conflicts.

The right question is not whether AI can “seize homes”

Saying that “AI seizes homes” is far too simple. Models have no legal authority, and data centres do not receive a special right to take land merely because they use AI. What exists is a chain of decisions: demand forecasts, grid planning, permits, negotiation with owners, state law and, in some cases, courts.

The useful public discussion keeps every link in view. AI expansion is making a familiar question more visible: who pays, where infrastructure is built and how a public benefit is defined. The answers will not come from a headline about inevitable takings. They will come from specific project records and public processes that are still open to scrutiny.