Data centers have become the face of America's AI infrastructure boom. But the real fight isn't about buildings. It's about who gets access to limited power, water, land, and public trust and who pays the cost of expanding them.
.svg)
Driving through rural Pennsylvania recently, I kept seeing the same sign.
No data centers in our neighborhood!
After a while, the signs started to feel less like opposition to a building and more like opposition to a bargain people never agreed to. A quiet road, a few farms, a township meeting, then suddenly the conversation is about substations, cooling systems, transmission upgrades, shell companies, tax breaks, and whether one facility might use as much power as a small city.
This is why the data center debate has spread so quickly. The public story is simple: data centers are coming for our water, our power, and our land.
That story is not entirely wrong. But it is incomplete.
The deeper issue is that data centers have become the most visible symbol of a much larger infrastructure problem. America wants AI, cloud computing, electric vehicles, heat pumps, semiconductor fabs, reshored manufacturing, and industrial electrification. All of that runs through the same bottleneck: land, power, water, permits, and local trust.
Data centers did not create that bottleneck. They are exposing it.
The strongest public fear is often water. That makes sense. Electricity is abstract. Water is personal. People understand wells, aquifers, drought, and the possibility that a company headquartered somewhere else could draw from a resource that feels local by nature.
But the national water picture looks different from the local political picture.
Lawrence Berkeley National Laboratory estimated that U.S. data centers consumed about 66 billion liters of water directly in 2023. Converted into gallons, that is roughly 17.4 billion gallons a year, or about 48 million gallons per day. That includes direct water used at data center sites, mostly for cooling. LBNL also estimates indirect water use from the electricity system, but that depends heavily on the grid mix supplying the facility.
Now compare that with how the United States uses water. USGS estimated total U.S. water withdrawals at 322 billion gallons per day in 2015. Irrigation alone was 118 billion gallons per day. Thermoelectric power was 133 billion gallons per day. Self-supplied industrial use was 14.8 billion gallons per day. Livestock withdrawals were 2 billion gallons per day.
The comparison is not perfectly apples to apples, because withdrawals and consumption are different measures. Still, it gives useful scale. Direct data center water consumption is tiny compared with irrigation, thermoelectric power, industrial withdrawals, and even livestock water use. EPA has also estimated that residential landscape irrigation uses nearly 9 billion gallons per day nationally, mostly for lawns and gardens. That is roughly 190 times the direct daily water use of all U.S. data centers in LBNL’s 2023 estimate.
Cattle are an especially useful comparison because the visible water use understates the system. Livestock withdrawals were 2 billion gallons per day in the USGS data, but USDA-linked research notes that beef and dairy production require large amounts of water primarily through irrigation of feed crops, and together account for about 9 percent of annual U.S. freshwater withdrawals when feed irrigation is included.
None of this means water concerns are fake. It means they are local problem.
A data center in a water-stressed aquifer using evaporative cooling is a very different community issue than a facility in a water-abundant region using closed-loop or dry cooling. The mistake in the public conversation is treating every data center as the same environmental object. Cooling technology, local hydrology, grid mix, peak-day demand, and water rights matter more than the category “data center” itself.
The electricity story is more serious. LBNL estimated that U.S. data centers used 176 terawatt-hours of electricity in 2023, or 4.4 percent of total U.S. electricity consumption. Its 2028 scenarios range from 325 to 580 terawatt-hours, equal to 6.7 percent to 12 percent of projected U.S. electricity consumption.
Let’s put that in context. Data centers are not the number one consumer of electricity in the United States. In 2023, electric utilities sold about 1,450 terawatt-hours to residential customers, 1,408 terawatt-hours to commercial customers, and 1,009 terawatt-hours to industrial customers. Data centers are usually embedded inside the commercial category, so the comparison is not perfectly clean, but the scale is clear: households, commercial buildings, and industry each consume far more electricity than data centers as a category.
Compared with manufacturing, data centers are large but not unprecedented. EIA’s 2022 Manufacturing Energy Consumption Survey estimated total manufacturing electricity net demand at about 933 terawatt-hours. Chemicals alone used about 210 terawatt-hours. Primary metals used about 104 terawatt-hours. Paper used about 97 terawatt-hours. Food manufacturing used about 87 terawatt-hours. Plastics and rubber products used about 71 terawatt-hours. Data centers, at 176 terawatt-hours in 2023, are smaller than chemicals and much smaller than manufacturing overall, but larger than many major manufacturing subsectors.
So the right electricity argument is not that data centers are America’s largest power user. The real issue is growth, concentration, and timing.
A chemical plant or steel mill is part of a slower industrial planning cycle. A hyperscale data center can arrive as a sudden block of demand in a place where the grid was not built for it. The facility may need new substations, transformers, transmission capacity, backup generation, or new power contracts. In some regions, the question is less “Can America generate enough electricity?” and more “Can this local grid absorb this much new demand without pushing costs onto everyone else?”
That is where public anger becomes rational. Residents may not know the exact terawatt-hour numbers. But they understand an unfair bargain when they see one. If a trillion-dollar company gets the power, the tax break, and the land, while households get higher bills and more infrastructure stress, the politics will turn ugly.
Pennsylvania’s response shows where this is heading. Governor Josh Shapiro’s 2026 executive order requires large data center developers seeking state permits to meet standards around energy affordability, transparency, community engagement, workforce development, and environmental protection. It also removes AI data center proposals from fast-track permitting and prohibits nondisclosure agreements for these projects.
That matters because the pushback is not only about emissions or gallons. It is about process.
People see anonymous land purchases. They hear about NDAs. They are told projects will create jobs, but many of the jobs are construction jobs, not permanent local employment. They are told the facilities are essential to the future, but the benefits feel national while the burdens feel local.
The conversation has focused on whether data centers are environmentally dangerous. The more useful question is whether communities are being offered a credible bargain.
A credible bargain would be simple. Developers disclose expected power use, water use, cooling methods, backup generation, noise impacts, and grid upgrade needs. They pay the full cost of new infrastructure required to serve them. They bring new power where needed. They meet local water standards. They do not hide behind shell companies or confidentiality agreements. They negotiate community benefits in public.
That would not eliminate opposition. Some communities still will not want data centers nearby. But it would move the debate from rumor to governance.
Data centers are not harmless. They use real electricity, real water, real land, and real grid capacity.
But they are not the largest environmental threat facing American communities. Agriculture, thermoelectric power, heavy industry, lawns, transportation, and legacy manufacturing all place larger claims on land, water, energy, and emissions.
The reason data centers feel so threatening is that they arrive fast, concentrate demand locally, and are often introduced through institutions people already distrust.
The challenge is not to convince communities that data centers have no impact. They do.
The challenge is to build a development model where the companies creating the new load also carry the cost of serving it.
Technology is not the story here. The story is the system around it: the grid, the permitting process, the water rules, the tax incentives, and the public trust required to build infrastructure in a country where every local community has learned to ask, reasonably, “What exactly are we being asked to give up?”
.png)
.png)
.png)