Most people have a vague sense that “the cloud” lives somewhere physical — that when you stream a movie, send an email, or ask an AI chatbot a question, the data doesn’t simply float through the air and land on your screen. It gets processed somewhere. That somewhere is a data center, and understanding what one actually is turns out to be essential to understanding why thousands of communities across America are now fighting to keep them out.
A data center is, at its most basic, a building — or more accurately, a campus of buildings — that houses the computer servers that power the modern internet. These aren’t personal computers. They’re industrial-grade machines, stacked in rows of metal racks, running 24 hours a day, 365 days a year, processing and storing the enormous volumes of data that digital life generates. Your bank’s transaction records live in a data center. So does your streaming service’s entire library. So does every email you’ve ever sent, every search you’ve ever run, and increasingly, every AI query you’ve ever typed.
Data centers come in several sizes. A small enterprise data center might occupy a single room in an office building. A medium-sized facility might take up a warehouse. But the ones driving the current national controversy — the ones communities are fighting — are hyperscale facilities operated by Amazon, Google, Microsoft, Meta, and a growing roster of AI companies. These are industrial campuses covering hundreds of acres, with buildings the size of aircraft hangars, surrounded by enormous electrical substations, cooling towers, and water storage tanks.
The typical large data center contains tens of thousands of servers. Each server generates heat — a lot of it. Managing that heat is the central engineering challenge of data center design and the source of the facility’s enormous resource demands. Left uncooled, servers fail within minutes. The entire operational model depends on maintaining a precise temperature range, continuously, at scale.
There are three primary cooling approaches in use today, and which one a data center uses determines its water footprint, its electricity demands, and its environmental impact profile.
The oldest and most common method. Cold air is pushed through the server racks, absorbs the heat they generate, and is then either exhausted outside or — in more efficient designs — routed through a chiller system that removes the heat before recirculating the air. Air-cooled data centers typically have a lower water footprint but higher electricity consumption, because the chillers themselves require significant power.
More efficient than air cooling, water cooling uses chilled water circulated through the facility to absorb server heat. The heated water is then routed to a cooling tower outside the building, where it is either evaporated into the atmosphere or passed through a heat exchanger. Evaporative cooling towers are the primary source of data centers’ notorious water consumption — the water that evaporates must be continuously replenished. A single large facility using evaporative cooling can consume millions of gallons per day.
An emerging technology where servers are submerged directly in a non-conductive liquid that absorbs heat far more efficiently than either air or water. Immersion cooling dramatically reduces both water consumption and electricity use, but its higher upfront cost means it remains rare in existing large-scale deployments. It represents a genuine technological path toward more sustainable data center operation — but it is not what is being built at scale today.
A hyperscale data center doesn’t just plug into the local electrical grid. It requires a dedicated substation — essentially a mini power plant on-site that steps down high-voltage transmission power to the levels the facility can use. These substations must be built specifically for each facility, often requiring new high-voltage transmission lines running for miles to connect the data center to the nearest point on the regional grid.
The electrical demand of a single large facility can be staggering. Data center power is measured in megawatts. Small facilities run on 1–10 MW. Hyperscale campuses routinely require 100–500 MW. The AI-era “gigawatt campus” — a single facility or cluster requiring 1,000 MW or more — is now being discussed in industry circles as the next frontier. For context, 1,000 MW is roughly the output of a nuclear power plant.
The major operators — Amazon Web Services, Google Cloud, Microsoft Azure, Meta, and Oracle — site their facilities based on a combination of factors: land cost, electricity cost and availability, water availability, fiber connectivity, tax incentives, and increasingly, political stability and regulatory permissiveness. States and counties compete aggressively to attract these facilities, offering tax abatements, infrastructure subsidies, and fast-tracked permits. The result is a concentration of facilities in a handful of regions where the conditions are favorable: Northern Virginia, the Phoenix metro, rural Indiana, the Reno-Sparks corridor, suburban Columbus, and the suburbs of Atlanta.
A newer category of operator has also emerged: specialized data center real estate investment trusts (REITs) and private equity-backed development companies that build “build-to-suit” facilities for tech tenants, or speculative campuses that they lease to multiple customers. These operators often have less public profile than the tech giants but are responsible for a significant share of new construction.
When data centers are proposed to local governments, they are almost always presented as job creation engines. The reality is more complicated. A hyperscale facility worth $1 billion or more typically employs 50–150 permanent workers. These are real jobs — and they pay well — but they represent a tiny fraction of what communities expect when they approve industrial campuses of this scale. The construction phase creates many more jobs, but those are temporary. Once the building is up, the permanent workforce is lean by design: data centers are engineered for automation.
Data centers have existed for decades without generating significant public controversy. What changed is scale and speed. The AI boom — driven by the explosive growth of large language models, image generators, and AI-powered services — has created a demand for compute power that is growing faster than anything the industry has previously experienced. Every AI interaction is computationally expensive. Every AI model trained requires enormous processing power over extended periods. The infrastructure to support this demand is being built right now, as fast as capital and permits allow, in communities that often have no meaningful say in the process.
That is the context for the fights happening in Virginia, Utah, Arizona, Texas, Nevada, Georgia, Indiana, and dozens of other states. The buildings going up in those communities aren’t ordinary warehouses. They are the physical substrate of a trillion-dollar industry’s rapid expansion — and the communities bearing the costs of that expansion are increasingly refusing to do so in silence.
“They said it would bring jobs. It brought 40 permanent positions and a substation that doubled our electricity rates.” — Resident of Loudoun County, Virginia, at a 2024 zoning hearing
The essays that follow go deeper on each dimension of this issue: the water, the power, the land, the environmental impacts, the legal landscape, and what communities that have successfully fought back actually did. Start with whichever issue matters most to your situation — or read them in sequence to build a complete picture of what you’re up against and what you can do about it.