Are Data Centers Destroying Communities?

by ChatGPT

There is a much more defensible middle ground than either “data centers are destroying communities” or “the concerns are nonsense.”

The evidence as of September 2026 says this: data centers are economically useful infrastructure and AI absolutely requires them, but the newest hyperscale AI centers are extraordinarily large industrial electrical loads. They create substantial construction activity and tax revenue, but relatively few permanent jobs. 

Their effects on electricity prices, water supplies, and neighbors depend heavily on where they are built and how they are engineered.

Here is how ChatGPT would separate fact from exaggeration.

Issue	What the evidence actually says
Electricity rates	Yes, data centers can raise regional electricity costs, particularly where new generation and transmission must be built rapidly. But good rate design can prevent ordinary customers from subsidizing data-center-specific infrastructure.
Permanent jobs	Not many for their physical size and investment. Construction employs many people temporarily; operation employs relatively few.
Noise	Usually not as loud as heavy manufacturing, but the continuous low-frequency hum can be unusually irritating to nearby residents. Poor siting is the main problem.
Water	Sometimes very large, sometimes surprisingly small. Cooling technology and climate dominate the answer.
Golf-course comparison	“A golf course uses twice as much water as a data center” is not a reliable generalization. It can be true for some air-cooled centers and very false for large evaporatively cooled hyperscale centers.
Are all these centers necessary?	More capacity is clearly needed, especially for AI, but nobody knows that every proposed project is necessary. Regulators have found speculative and duplicate projects in utility connection queues.

Electricity: this is the strongest legitimate concern

The latest Lawrence Berkeley National Laboratory estimate, published in June 2026, puts U.S. data-center electricity consumption at a central estimate of about 11.8% of all U.S. electricity by 2030, with a plausible range of 9.5% to 15.3%. That is an enormous change from an industry that consumed only about 4% several years ago. (BIES Division)

The International Energy Agency finds that data centers are likely to account for roughly half of U.S. electricity-demand growth through 2030. AI is the fastest-growing component. (IEA)

There is real evidence that this can affect prices. Virginia commissioned an independent analysis because it has the world's largest concentration of data centers. It estimated that data-center-driven generation and transmission expansion could add roughly $14 to $37 per month by 2040 to the generation and transmission portion of a typical Dominion residential customer's bill, depending on how much development actually occurs. (JLARC)

PJM, the grid operator serving 13 states and Washington, D.C., says data centers are now the primary source of its projected load growth. It also says that demand has been growing faster than new generating capacity, contributing to higher capacity prices. That does not mean data centers caused the entire increase—generator retirements and other factors matter too—but they are clearly a major factor. (PJM Inside Lines)

There is an important counterpoint. Virginia's investigation found that existing rates were generally allocating existing costs appropriately to data centers. The danger comes primarily from the massive amount of new infrastructure required and the risk that utilities build it for projects that later disappear. (JLARC)

That is why Virginia has created a special large-load rate class requiring major customers such as data centers to make long-term commitments and pay minimum transmission and distribution charges. This is a sensible way of protecting residential customers. (Virginia SCC)

So the statement “data centers raise everybody's electric bill” is too broad. A better statement is:

Rapid data-center growth can raise regional electricity prices when it forces expensive generation and transmission expansion. Whether residential customers bear those costs depends substantially on utility regulation and rate design.

Jobs: the claims are often exaggerated

This is one of the weakest arguments made in favor of data centers.

Virginia's legislative research agency found that a typical 250,000-square-foot data center employs only about 50 full-time workers, approximately half of whom may be contractors.

During construction, however, employment is substantial. A project may have approximately 1,500 workers on site at the peak of its 12- to 18-month construction period. (jlarc.virginia.gov)

The long-term economic attraction is therefore not thousands of operating jobs. It is enormous capital investment and potentially substantial property and equipment taxes. Virginia found that in established data-center communities, data centers contributed anywhere from less than 1% to as much as 31% of local government revenue, depending upon local tax policy and concentration. (jlarc.virginia.gov)

Thus:

Data centers can be excellent tax-base developments, but they are poor choices if a community's principal objective is large numbers of permanent jobs.

Noise: different from a factory rather than necessarily louder

Calling a data center “noisier than a factory” isn't particularly meaningful because factories vary enormously.

A steel mill, stamping plant or sawmill can obviously be much louder than a data center. A warehouse can be quieter.

Data-center noise comes principally from cooling fans, chillers, transformers and occasionally backup generators. The distinctive issue is that cooling equipment may operate 24 hours a day and can produce persistent low-frequency sound.

Virginia's extensive investigation found that the sound usually was not loud enough to damage hearing and rarely violated conventional noise ordinances. Most data centers generated no noise complaints at all. But where they were placed too close to residential neighborhoods, some residents reported significant problems from the continuous low-frequency sound. (jlarc.virginia.gov)

That makes this primarily a zoning and engineering problem. Several hundred yards of industrial separation, sound modeling, barriers and appropriate equipment design can make an enormous difference.

I would therefore describe a well-designed data center as generally less disruptive than heavy manufacturing but potentially more annoying than its decibel reading suggests because it never shuts off.

Water: this is where statistics become especially misleading

U.S. data centers consumed about 66 billion liters—17.4 billion gallons—of water directly in 2023, primarily for cooling. That was about triple their 2014 consumption. (PLOS)

But individual facilities are wildly different.

Some use evaporative cooling and consume hundreds of thousands or even millions of gallons per day. Others rely primarily on air cooling or closed-loop liquid cooling and consume comparatively little fresh water.

Microsoft, for example, reports that its newer closed-loop design eliminates water evaporation for cooling; it estimates that the design can avoid more than 125 million liters—about 33 million gallons—of water annually per data center compared with its previous design. (Microsoft)

There is also an important distinction between onsite water and water used elsewhere to produce electricity. Berkeley Lab estimates that the electricity supplying U.S. data centers was associated with roughly 800 billion liters of indirect water consumption in 2023. Those numbers should not simply be combined, because that water is consumed at power plants in different places and depends enormously on the generating technology. (PLOS)

What about that golf-course claim?

This one interested me because it is a good example of how technically true comparisons can become misleading.

The Golf Course Superintendents Association reports that the median U.S. golf facility used 68.9 acre-feet of irrigation water in 2024. That works out to approximately 22.5 million gallons per year, or about 61,500 gallons per day averaged over a year. (GCSAA)

Nationwide, golf facilities applied about 1.63 million acre-feet of water in 2024. (GCSAA)

Compare that with a large AI-oriented hyperscale facility. A recent research assessment estimates roughly 2 million liters—about 528,000 gallons—per day for an average 100-MW hyperscale data center using conventional water-consuming cooling. (Drexel University)

So in that comparison, the data center uses roughly eight times the median golf facility—not half as much.

But compare the same golf course with a data center using primarily air cooling or Microsoft's new zero-evaporation cooling system, and the golf course could indeed consume much more water.

Therefore I would not repeat the “golf courses use twice as much water” claim. There is simply too much variation in data-center design for it to be meaningful.

One broader perspective is useful, though: total U.S. golf irrigation water consumption remains vastly larger than direct data-center water consumption nationally. That does not tell you whether a particular data center is a problem. A 500,000-gallon-per-day facility in a water-stressed community matters considerably more than several million gallons somewhere with abundant water.

Do we really need this many data centers for AI?

We definitely need substantially more computing capacity if current AI usage continues.

The IEA reported in April 2026 that electricity use by AI-focused data centers increased about 50% during 2025 and projects their electricity consumption to roughly triple between 2025 and 2030. At the same time, the energy required for individual AI tasks is falling rapidly because chips, models and software are becoming more efficient. (IEA)

So there is a genuine computing boom—not an invented one.

But that does not establish that every announced data center is necessary.

FERC has specifically identified speculative data-center connection requests. Companies sometimes submit requests at several potential sites while deciding where to build. FERC warns that this can cause double counting, inflate utility demand forecasts and potentially lead to unnecessary infrastructure investment. (Federal Energy Regulatory Commission)

There is another uncertainty: efficiency. The IEA estimates that stronger hardware, software and infrastructure efficiency could provide the same digital services with more than 15% less data-center electricity consumption than its base projection. (IEA)

Therefore I would not accept either of these arguments:

“Every proposed AI data center is needed.”

or

“We don't need all these data centers because AI will become more efficient.”

The evidence supports neither. Demand is growing extremely rapidly while efficiency is also improving extremely rapidly. Which wins over the next decade remains uncertain.

The real pros and cons

The substantial benefits are real: data centers provide the computing infrastructure behind cloud computing, communications, banking, research and AI; they bring enormous private capital investment; they produce major temporary construction employment; they can generate unusually large property-tax revenues; and major technology companies can help finance new generation, transmission, nuclear, geothermal, renewable power and storage.

The substantial disadvantages are equally real: hyperscale centers consume extraordinary amounts of electricity; concentrated development can force billions of dollars of generation and transmission investment; poorly designed tariffs can shift some costs to other customers; permanent employment is modest; some cooling systems consume substantial water; poorly sited facilities can create persistent noise and visual impacts; and governments sometimes grant very large tax incentives to attract facilities that would otherwise produce enormous tax revenue.

The issue, therefore, is not really “Are data centers good or bad?”

The more useful question is:

Under what conditions should a community permit one?

Based on the evidence, I would examine five things very closely: who pays for the electric infrastructure, how much water the particular cooling design actually consumes, distance from residences, the long-term tax revenue after incentives, and whether the developer has made a firm long-term electrical-load commitment rather than merely reserving capacity.

If those five questions are answered satisfactorily, many of the strongest objections to a data center become much less compelling. If they are not, a community can inherit significant costs for surprisingly little permanent employment.