How Many Data Centers Are in the US, and How Many Are Planned?

Ask five sources how many data centers are operating in the United States and you get five different answers. These are all published 2026 figures, for the same country, in the same year.

There is also no credible year by year count of US data centers. No directory publishes a consistent back series, and the federal data cannot supply one, so any chart showing facilities per year since 2014 is built on a definition that changed underneath it. The only defensible annual series for this industry is electricity, which is why this page charts that instead.

The gap between the counts above is not sloppiness. It is definitional. A directory that counts commercial colocation facilities will list a single suite inside a downtown office building as one data center. A directory that counts campuses will record a 500 acre hyperscale site with six buildings as one data center too. Some include enterprise server rooms run by hospitals and banks. Some include crypto mining facilities; some do not. Some count announced projects alongside operating ones.

There is no licensing regime that would settle it. A data center is not a regulated category the way a landfill, a hospital, or a power plant is. In most of the country it is approved as an ordinary industrial or commercial use, sometimes by right, which is precisely why so many residents discover one is coming only when the site work starts.

The federal government has one annual series that touches this industry: the Census Bureau's County Business Patterns, which counts business establishments under NAICS 5182, retitled in 2022 as Computing Infrastructure Providers, Data Processing, Web Hosting, and Related Services. We pulled the national and state files directly for 2017 through 2023.

It does not work, and one comparison shows why.

Northern Virginia is the largest data center market in the world, roughly 20 gigawatts of capacity and on the order of 13 percent of all live global capacity, with Loudoun County alone holding more than 30 million square feet. The federal series has Virginia sixth and shrinking. California leads it, because California is where hosting and software companies keep their offices.

County Business Patterns counts corporate headquarters, not server halls. A hyperscale campus is usually owned by a company registered in another state, so the building lands in your county while the establishment is recorded somewhere else, or is absorbed into a parent filing under a different code. Nationally the count barely moved, rising about 5 percent between 2017 and 2023, across the exact years this industry became a national controversy. The buildings are real. They are simply not in this table, which is why every number in circulation comes from a private directory instead.

If you want a defensible national trend, stop counting buildings and count power. Lawrence Berkeley National Laboratory, funded by the Department of Energy, published the 2024 United States Data Center Energy Usage Report in December 2024. It is the standard reference, and it is the number that actually governs what happens in your county, because power is what a data center competes with you for.

Two details in that report matter more than the headline. First, growth accelerated sharply: the compound annual growth rate was about 7 percent from 2014 to 2018 and about 18 percent from 2018 to 2023. Second, the 2028 figure is a range built from scenarios, not a prediction. The distance between 325 and 580 terawatt hours is mostly a question of how many AI accelerators ship and how hard they are run. Anyone citing a single 2028 number has picked one end of a band and dropped the other.

We chart only the years the report states in its own text. It presents the full series as a figure rather than a table, and we do not publish numbers read off a chart as though they were tabulated data.

Pew Research Center analyzed the Data Center Map database as of February 19, 2026 and found more than 1,500 US data centers in development against roughly 3,000 operating. The geography of that pipeline is the single most important fact on this page for anyone reading it from home.

Project counts understate the change, because the projects being built now are far larger than the ones already standing. Measured in power, which is the unit that decides whether a project strains your grid and your water, the pipeline dwarfs the installed base.

Here is the part that almost no coverage explains. The enormous numbers you see quoted for future data center power are drawn from utility interconnection queues, and a queue is a list of requests, not a list of projects. Filing costs a developer very little. Not filing, in a market where grid capacity is finite and first in line wins, could cost them everything. So developers file for the same project with several utilities in several states, then build in one place and let the other applications sit until they are withdrawn.

The result is double counting on a scale that distorts every forecast built on it.

Utility planners say this openly. Zachary Ruzycki, director of resource planning at Great River Energy in Minnesota, told Utility Dive in May 2025 that "Some requests could be duplicates, but there's no good way to tell which." In the same reporting, Camus Energy chief executive Astrid Atkinson estimated that speculative interconnection requests ran five to 10 times higher than the number of data centers actually being built. Near Minneapolis and St. Paul, at least 11 data center campuses had been proposed since 2020 and only one, Meta's, had begun construction.

When a utility applies real filters, the number collapses. Exelon reported roughly 11 gigawatts of what it classified as high probability data center load over a decade, a small fraction of what its queues contained.

The statistic that should shape your expectations is Pew's: 39 percent of planned data centers are going into counties that have none today. Most communities now facing one have no local precedent, and very often no ordinance written with this use in mind. That cuts both ways. It means your county probably has no noise standard tailored to a facility that runs every hour of every day, and it also means the rules are still being written, which is the only period when a resident has real leverage.

Three things follow from the data on this page.

Early is the only time the numbers are soft. Before a rezoning or a special use permit is granted, conditions can still be attached: setbacks, decibel limits with a stated measurement method, water commitments, road agreements, and decommissioning bonds. After approval, those are contract terms you were never a party to. Our guide on how to fight a proposed data center walks the sequence, and some communities have used a moratorium to buy the time to write real standards.

The power request tells you the true size. A project described as a few buildings may have asked the utility for hundreds of megawatts. That figure predicts noise, water use for cooling, and pressure on your rates better than the site plan does. Start with who pays for data center power and what it does to your electric bill.

If one is already running near you, document now. Whatever the national count turns out to be, your claim depends on your own record: dated decibel readings, well tests, photographs, and complaint logs. Our guide on the evidence to document data center harm is the place to start, and if you want a lawyer to look at it, a free case review connects you with independent attorneys handling these cases in your state.