Demand for computing capacity has driven a building programme at a scale the industry has not seen, and the effects are becoming visible locally in a way they previously were not.
The scale
A large facility draws power comparable to a small city, continuously. Connecting that to a grid designed for ordinary industrial load is the main constraint on where they can be built, and the queue for connections runs to years in many places.
The local objections
Electricity prices, where communities absorbing large new load have seen upward pressure on residential bills.
Water, since many cooling systems consume substantial quantities, which is contentious in dry regions.
Noise, from cooling equipment running continuously, which is the most common complaint from people living nearby.
And employment, which is modest relative to the investment – these facilities are automated and employ far fewer people than the capital suggests.
What operators point to
Tax revenue, which is substantial. Investment in grid infrastructure that also serves the area. And long-term power purchase agreements that fund new generation, including nuclear in several recent cases.
Where this is heading
More scrutiny at the planning stage, which is already slowing approvals in several jurisdictions.
Location decisions driven by power availability rather than proximity to users, which works for training workloads and not for serving responses.
And more efficient cooling, because water use has become the most politically difficult element.
Why it matters
The constraint on AI capability over the next few years is less likely to be research than a substation with a five-year upgrade queue.