The Towns That Power the Machines
An NTT Data analysis warns that electricity, land and permitting now decide where AI data centers get built — and which towns quietly bear the cost.
The AI boom is quietly turning into a fight over electricity, land and the towns chosen to host the machines. An analysis from NTT Data warns that access to power — not chips, not code — is becoming the deciding factor in where new AI data centers can be built and how fast. The uncomfortable question underneath the growth curve is who bears the cost of intelligence, and the answer is increasingly a place most people have never heard of.
Somewhere tonight, a county commissioner is looking at a site plan for a building the size of several football fields, windowless, humming, promising jobs and tax revenue and a draw on the grid that rivals a small city. The pitch is familiar and seductive: the future wants to live here. What the pitch says less loudly is that the future is thirsty. GPU-dense clusters, the kind that train and run large models, pull power and cooling at a scale that older data centers never approached.
Why electricity became the bottleneck
For years the story of AI was told in silicon — faster chips, bigger models, the race for GPUs. The report from NTT Data reframes it as an infrastructure story, and that shift matters. When electricity becomes the constraint, the map of who gets to build changes overnight. Regions with spare generation capacity and permissive permitting move to the front of the line. Regions without it wait, or watch the projects go elsewhere.
According to the original NTT Data analysis, the strain is not limited to power. It stacks — cooling, specialized equipment, land, and the slow machinery of permitting all pile on top of the same demand curve. Each of those is a place where a project can stall, and each is a place where a community can push back or be pushed aside.
The number worth sitting with is this: a single large AI campus can draw as much electricity as a mid-sized town, continuously, for years. That is not a spike. It is a permanent new resident that never sleeps and never leaves, competing for the same grid capacity that heats homes and runs hospitals. When a utility approves that connection, it is making a quiet decision about priorities that most ratepayers never got to vote on.
Who actually bears the cost?
Here is the part that rarely makes the announcement. The benefits of an AI data center — the models, the products, the market valuations — accrue globally and abstractly. The costs land locally and concretely: on the water table, on the electrical bill, on the two-lane road now carrying construction traffic, on the ridge line that used to be dark at night. Intelligence, as an industry, is very good at externalizing its footprint.
I keep thinking about the asymmetry of it. A family in one of these host towns may never use the model their local grid is now feeding. They will, however, feel the effects of the choice — in rates, in land use, in the character of a place that agreed to become infrastructure. The word we use for this exchange is "investment," but investment implies the returns come back to the people who put something in. Often they do not.
None of this makes data centers villains. The machines are useful, sometimes profoundly so. The point is narrower and harder: the AI era is being built out of real electricity, in real places, on the backs of real permitting fights, and the people making the decisions are frequently not the people who will absorb the consequences.
What towns should ask before they say yes
The communities getting these proposals deserve better questions than "how many jobs?" They should ask what the campus draws at full load, where the water for cooling comes from, who pays for grid upgrades, and what happens to the rates of everyone else on the line. They should ask what the site becomes if the model economy cools and the servers go dark. A boom that treats towns as disposable substrate is not a boom they should welcome uncritically.
The deeper shift is philosophical. We have started to talk about AI as though it were weightless — a mind in the cloud, everywhere and nowhere. The truth is heavier. Every answer has an address. And the story of the next decade may be less about what the machines can think and more about which towns we decided should power the thinking.