PERSPECTIVE | AI, ECONOMIC DEVELOPMENT AND WORKFORCE READINESS
Perspective No. 5 · Economic Development
Living With the Buildout, and Living Without It
AI Data Centers, Community, and the Environment
Download this Perspective (PDF)Every week brings another announcement. A company will invest billions in a new data center campus. Jobs will be created. A region will become a hub for artificial intelligence. And within days, the same argument starts, in the same order, almost word for word, in a place that has never hosted one of these facilities before.
One side talks about the future. The other side talks about the electric bill. Both are describing the same building.
There is also a second kind of place in this story, and it gets less attention: the place the announcement did not name. The county that was considered and passed over. The town that debated a facility and decided against it. The region that was never in contention at all. Those places are living with the buildout too, from the other side of the line, and any honest account of this moment has to include them, because the buildout does not distribute its effects only to the places that host it.
That is the actual state of the debate over AI infrastructure, and it is why the debate so rarely settles anything. The two sides are not disagreeing about values. They are answering different questions. One is answering a question about capability: what this technology makes possible, and what it costs a place to be left out of it. The other is answering a question about carrying costs: what this facility will use, who pays for the upgrades it requires, and what remains in the town when the construction crews leave. Until both questions are on the same page, the argument will keep restarting with every announcement.
The numbers exist, published by the institutions that measure this. The questions host communities ask are remarkably consistent from one country to the next. And the places that answered those questions early have a record worth looking at. The pattern is not the one either side of the argument usually describes.
Small Globally. Decisive Locally.
One step back first. Nobody in this debate is choosing whether to use data centers. Everyone already does, mostly without knowing it. The photos backing up from a phone, the medical record a doctor pulls up mid-appointment, the card payment at a register, the school platform a child logs into at night, the video call with family three states away: all of it runs in one of these buildings. Data centers have quietly become part of the basic equipment of growth, in the same category as the power grid and the road network. The argument is never really about whether they exist. It is about where they sit, what they draw from the places that host them, and what those places get in return.
Start with scale, because scale is where the public argument goes wrong after that. The International Energy Agency, in its 2025 report Energy and AI, estimates that data centers consumed about 415 terawatt-hours of electricity in 2024, roughly 1.5 percent of the world's electricity. Under the agency's Base Case, that figure more than doubles by 2030, to about 945 terawatt-hours, roughly what Japan consumes today. Those are the numbers behind the headlines, and they are real.
But nobody lives in the global average. A data center does not draw 1.5 percent of a planet's power. It draws from one grid, one substation, one watershed, in one place. In the United States, Lawrence Berkeley National Laboratory estimates that data centers used 176 terawatt-hours in 2023, about 4.4 percent of national electricity, and projects a range of 325 to 580 terawatt-hours by 2028, between 6.7 and 12 percent of the total. In Ireland, the Central Statistics Office reports that data centers consumed 22 percent of all metered electricity in 2024, up from 5 percent in 2015, and more than the country's urban households combined. The global figure says the buildout is manageable. The local figures say that in the places where the buildout concentrates, it is one of the largest planning facts in public life. Both statements are true. The argument about the buildout is usually conducted in the first set of numbers. The decisions get made in the second.
Water follows the same shape. Berkeley Lab estimates put direct water consumption by US data centers at about 66 billion liters in 2023. As a national total, that is a modest figure. As a draw on one aquifer or one municipal system in a dry year, it can be the largest single fact in a county's water plan. How much water a facility uses is also a design decision, not a fixed property: cooling systems differ, and the choices an operator makes at the design stage determine most of the footprint for the life of the building. Which is why the water question, like the power question, belongs at the beginning of a project, when the answers can still change the design.
The Four Questions
Strip the hearings and the headlines down, and host communities ask the same four questions, in Ireland and Singapore and the American Midwest alike.
The first is about power. Who pays for the generation and the grid upgrades this facility requires, and what happens to the electricity bills of the households and businesses that were already there? The second is about water. How much will the facility use, from which source, and what happens in a dry year? The third is about what remains. Construction employment is temporary by definition. How many permanent jobs does the facility carry, who locally is prepared for them, and what does the site contribute to the tax base over decades, not ribbon cuttings? The fourth is about voice. Were the people who will live next to this facility asked before the announcement, or informed after it?
None of these questions is an argument against building. They are the questions any prudent community asks about any large industrial neighbor, and they have the same property everywhere: they are answerable. A project either knows its water source or it does not. Its effect on local rates is either modeled and published or it is not. Its permanent workforce is either a number in writing or it is a rendering. The communities asking these questions are not asking for the buildout to stop. They are asking for the answers to be part of the consideration from the start, while a project can still be shaped by them.
Both Sides of the Fence Line
What does living with the buildout actually look like, in economic terms broad enough to describe most places that host it? Construction first: a large facility is years of work for electricians, lineworkers, HVAC technicians, welders, and the local firms that supply and service a major build, and demand on that scale is what lets a region sustain apprenticeships and training pipelines instead of running them cohort by cohort. Operations second: a smaller permanent workforce than the construction peak, but a durable one, surrounded by maintenance, security, logistics, and supplier relationships that recur every year the building runs. And underneath both, a tax base effect: a facility of this size becomes one of the largest ratepayers in its jurisdiction, and the grid and road improvements built to serve it are frequently usable by the employers who come after. None of that is automatic, and none of it should be promised in advance. It is what the with side of the ledger can contain when a project is built and run well.
There is a longer horizon on the with side, and it belongs to the workforce rather than the facility. The skills a buildout demands, electrical work, cooling and mechanical systems, network operations, safety, project management, are not skills for one building. They are the same skills that advanced manufacturing, hospitals, logistics, and grid modernization are short of everywhere. A region that builds the training ecosystem for a data center has built it for every employer who follows, and a worker who gains those skills carries them for a career, across employers and across industries. That transferability matters on the without side too. The ecosystem can be built before any facility is ever considered, as a region's own investment in its people, and it keeps its value if the facility never comes. Of everything the buildout touches, the skills are the part that stays, and the part that moves with the people who hold them.
The without side of the ledger is real too, and it is not one thing. A community that weighs a facility and declines it has made a genuine decision, and it keeps genuine things: its water, its grid headroom, its land, its quiet. Some places make that trade with open eyes and would make it again. But the capacity does not disappear because one place declines it. It is built somewhere else, and the trades work, the permanent jobs, and the tax base go with it. A place can also lose the buildout without ever being asked: no substation capacity, no fiber, no workforce within reach, and the announcement simply names somewhere else. For those places, living without the buildout is not a choice they made. It is a condition they are in, while the AI economy arrives anyway, in the services their residents use, the prices their employers pay, and the advantages accumulating to regions that were ready.
That symmetry is the part of the debate that gets lost. Hosting carries costs that deserve honest accounting. So does absence. A community on either side of the line is making a decision about its economic future, and it deserves the full ledger, both columns, before it decides.
What Places Have Done With Those Questions
The jurisdictions that moved first did not ban the buildout. They conditioned it, and the record of what happened next is the most useful evidence in this debate.
Ireland, where the share of national electricity going to data centers passed a fifth, moved to case-by-case scrutiny of new connections around Dublin, with the grid operator weighing each large request against system capacity. The Netherlands restricted hyperscale facilities to designated areas under national direction. Singapore paused new data center development for roughly three years, then reopened it through competitive calls in which capacity is allocated to operators who meet published standards for energy efficiency and green energy. Its 2024 Green Data Centre Roadmap provides at least 300 megawatts of additional capacity on those terms, with more available to projects powered by green energy. Building continued in all three places. What changed was the sequence: the conditions were published first, and projects were designed to meet them, instead of the conditions being negotiated in public after an announcement had already hardened into a fight.
The United States is arriving at the same place by a more local road: utility commissions testing how large-load customers should pay for the grid they require, and host communities negotiating conditions project by project. The direction is consistent across all of it. Support for the technology itself remains broad. What has ended, in place after place, is the assumption that a project should be approved first and explained later.
The Builder's Side of the Table
A factual account has to include the view from the other chair. A data center is among the most capital-intensive buildings a company can construct, built on timelines measured in years, into grid interconnection queues that are already long. Uncertainty is a genuine cost in that arithmetic, and an approval process whose rules change mid-project, or whose standards are invented in a public fight, is a cost that lands on everyone, including the community waiting for the tax base.
This is why a published conditions framework is not a concession extracted from builders. It is the one arrangement that serves every party at the table. A serious operator can design to a published standard: this much efficiency, this water accounting, this share of grid costs, this local workforce commitment. A community can judge every proposal against the same questions instead of relitigating first principles on every application. And a public official can point to a standard that existed before any particular project arrived, which is the only durable protection any of them has. The projects most exposed in the current environment are not the ones facing conditions. They are the ones whose sponsors treated the four questions as a communications problem, and discovered that a community's questions do not go away because a press release declines to answer them.
None of this asks an operator to open its books or publish its business. The ask is smaller, and more human, than the argument usually makes it sound: make the attempt, visibly and early, to be a good partner. Share what a community genuinely needs to plan around. Answer the four questions in good faith, including the ones with uncomfortable answers. Stay in the conversation after the announcement, when the questions turn to substations and school taxes. Operators who work that way exist in every market, and their projects tend to move. The ones who treat a host community as an obstacle to be managed tend to meet the version of that community that behaves like one.
Answer on Paper First
The debate between AI, data centers, the community, and the environment is usually described as a contest between the future and the place it lands in. The numbers do not support that framing. Globally, this buildout is a manageable share of a growing system. Locally, it is large enough to deserve real answers. Both things are true at once, and the places that have done best by their residents and by the industry are the ones that acted on both: they published the conditions, allocated the capacity, and let projects proceed that could meet the standard in writing.
None of that requires anyone to win an argument. It requires the answers to be on paper, where a community can weigh them, early enough to matter. The buildout will be measured twice in the end: once in megawatts, and once in whether the places that host it would agree to host it again.
---
TRANSLATION
| Policy priority | Practical first action | Evidence of progress |
|---|---|---|
| Grid costs, in the open | A host community asks to see the interconnection and grid-upgrade cost allocation as part of how it considers the project | Allocation shared in the public record |
| Household bills, addressed directly | The utility regulator's review of large-load rate design, with the effect on existing customers addressed in writing as the project is considered | Regulator's finding published as part of the process |
| Water, accounted like power | A shared projection of water use by source, including a dry-year plan, with actual use reported annually against it | Projection public; annual reported use within the projected range |
| What remains after construction | Permanent employment, local workforce preparation, and long-term tax contribution discussed as part of the community's consideration, not only in the announcement | Stated in writing; revisited as the project develops |
| The other side of the line | A region not hosting facilities states its own position honestly: what it keeps, what it gives up, and what would have to change for it to compete | A published local assessment, revisited as conditions change |
Living with this buildout and living without it are both real conditions, with real costs on each side. The honest conversation starts by admitting both.
Sources
- International Energy Agency, Energy and AI, April 2025 (global data center electricity use, 2024 estimate and 2030 Base Case).
- Lawrence Berkeley National Laboratory, 2024 United States Data Center Energy Usage Report, December 2024, released by the U.S. Department of Energy (U.S. electricity use 2023, projections to 2028, direct water use estimate).
- Central Statistics Office, Ireland, Data Centres Metered Electricity Consumption (data center share of metered electricity, 2024; urban household share).
- Infocomm Media Development Authority, Singapore, Green Data Centre Roadmap, May 2024 (additional capacity of at least 300 MW under efficiency and green energy criteria).

