The question gets asked every time a new campus breaks ground: why here? Why is Arizona, of all places, attracting the kind of buildout that used to land in Northern Virginia or the Bay Area? Metro Phoenix now ranks #2 in North America for planned data center development, behind only Northern Virginia. The casual answer for why is "tax incentives," and the casual answer is wrong. Tax incentives matter, but every state has them. They aren't decisive.

What's decisive is geography, and Arizona happens to sit at the intersection of four properties that data center developers actively look for. Most places have one or two. Almost nowhere has all four.

The Disaster Profile

Start with what Arizona doesn't have. No hurricanes. No tornadoes worth planning for. No earthquakes above the level that ordinary commercial construction handles. No flood plains that a competent siting team can't avoid. No wildfire risk that meaningfully threatens an urban or near-urban industrial site. No volcanic activity. No tsunami exposure. No ice storms that take out transmission lines.

This matters because data centers are designed for continuous operation. The standard reliability target is 99.999 percent uptime, which works out to roughly five minutes of unplanned outage per year. The dominant cause of large-scale outages, historically, is weather. A site where the disaster portfolio is essentially blank is a site where the engineering team isn't designing around contingencies that other regions can't avoid.

Compare to Houston, where hurricane season dictates building hardening and generator runtime planning. Compare to Northern California, where seismic codes drive structural cost. Compare to Atlanta, where ice storms and tornado tracks both factor into design. Phoenix's threat model is closer to "an unusually hot summer" than to "an act of God." That's an enormous engineering simplification, and it shows up in capex, opex, and insurance.

The Grid Topology

Arizona is on the Western Interconnection, one of the two major synchronized AC grids in North America, covering most of the western United States and parts of Canada and Mexico. The Western Interconnection has a different structure than the Eastern Interconnection or ERCOT in Texas. It has long transmission distances, fewer congestion points in the southern leg, and a generation mix that includes substantial nuclear baseload (Palo Verde), coal that's transitioning out, natural gas that's flexible, and rapidly growing solar.

For a data center, the interesting attribute is that Arizona has spare transmission capacity in the right places. Palo Verde Generating Station, which produces more electricity than any other power plant in the country, sits about 45 miles west of downtown Phoenix and feeds the West Valley directly. Its three reactors run on treated effluent rather than a lake or river, the only nuclear plant in the world cooled that way, and its 500-kilovolt switchyard is a pricing reference point for the whole Southwest grid. The grid was built to move that power, and the path it takes runs through the same corridors where data centers are siting today.

The siting overlap isn't coincidence. Developers look at where existing 230-kilovolt and 500-kilovolt lines run, where existing substations have headroom or can be expanded, and where utility planning documents indicate near-term capacity additions. The West Valley, the East Valley, and Pinal County all have this profile. Goodyear, El Mirage, Mesa, Chandler, Queen Creek, Casa Grande. The map of new data center campuses tracks the map of grid-favorable sites with surprising precision.

Compare to Northern Virginia, where Loudoun County's grid has been at or near capacity for years and Dominion Energy's interconnection queue has multi-year delays. The Phoenix metro has friction, but it isn't the same kind of friction.

The Fiber Geography

Long-haul fiber follows railroads, highways, and a small number of dedicated builds. The major east-west routes connecting Los Angeles to Dallas pass through Phoenix. The major north-south routes connecting the Pacific Northwest to Mexico pass through Phoenix. The major Southwest backbone fiber that serves Las Vegas, Albuquerque, El Paso, and Tucson passes through Phoenix.

That convergence is unusual. Most second-tier metros are on one or two backbones. Phoenix is on four or five. The redundancy means that a data center customer can buy capacity from multiple carriers on multiple physical routes, which is a hard requirement for hyperscale tenants and a significant cost advantage even for smaller users. It also means that latency to the major Southern California exchange points (where most transpacific cable lands) is in the 10 to 12 millisecond range, and latency to Dallas is in the 18 to 20 millisecond range. Those numbers matter for content delivery, AI inference, and database replication.

The fiber wasn't laid here because data centers needed it. It was laid here because Phoenix is where every Southwestern transportation corridor converges. Data centers showed up later, found the existing routes, and built campuses near the meet-me rooms.

The Land

Industrial land is the underrated variable. A hyperscale campus needs 100 to 500 contiguous acres, ideally flat, ideally close to a substation, ideally with truck access, ideally without endangered species or cultural resource issues, and ideally permittable in months rather than years.

Arizona has this. The West Valley alone has thousands of acres of permitted industrial land that wasn't already absorbed by manufacturing, distribution, or residential development. Pinal County has more. The Sonoran Desert is, from a developer's standpoint, a flat industrial substrate with mature water rights, mature transportation infrastructure, mature electrical infrastructure, and minimal environmental friction once the standard ESA and NHPA reviews are cleared.

This is partly geography and partly history. The Salt River Project, the Roosevelt Dam, the canal system, and the post-WWII manufacturing build (Motorola, then later Intel, then later TSMC) created an industrial corridor that mature industries could plug into. Each generation of arrival added infrastructure that the next generation could use. Data centers are the latest arrival, not the first.

Texas has comparable land. So does parts of the Mountain West. But Texas trades land for grid friction (ERCOT), and the Mountain West trades land for fiber friction (long distances to major exchange points). Arizona has all three.

The Climate Argument, Inverted

The conventional argument is that Phoenix is a bad place for data centers because it's hot. Hot ambient air means harder cooling, which means more energy spent on cooling, which means a worse Power Usage Effectiveness number.

This is true and it is also less significant than it looks. Cooling is a function of two variables: how hot the outside air is, and how much temperature differential the cooling system needs to reject. Modern data center designs operate cold-aisle temperatures in the 75 to 80 degrees Fahrenheit range, not the 65 degrees the industry used to target. A Phoenix summer afternoon, where ambient hits 110, has a delta of 30 to 35 degrees. A Northern Virginia summer afternoon, where ambient hits 95, has a delta of 15 to 20 degrees. The Phoenix delta is larger, but it isn't catastrophic, and the modern designs have closed most of the gap.

Phoenix also has free-cooling hours that Virginia doesn't have: dry winter nights where the outside air can do most of the cooling work without active refrigeration. Total annual cooling energy in Phoenix is higher than Virginia, but the gap is in single-digit percentages, not the 30 to 50 percent that critics often assume.

The bigger climate factor is humidity. Dry air is easier to cool with evaporative methods, which is part of why so many older Phoenix data center designs used water-intensive cooling. As the industry transitions to liquid cooling, humidity matters less, and the climate disadvantage shrinks further.

The Boring Answer

Phoenix won the data center lottery because the state has been quietly building industrial infrastructure for 70 years and the geography happens to be friendlier to large-scale buildings than most people realize. There isn't a single dramatic reason. There's a stack of small advantages that compound, and the developers who run the spreadsheets have noticed.

The compounding part is what's interesting. Once a region attracts hyperscale tenants, the supply chain follows: mechanical contractors, electrical contractors, controls integrators, generator service companies, switchgear distributors, fire protection specialists, security firms, fiber splicers, water treatment vendors. Phoenix has now built the full supply chain. The next campus is easier than the last campus, and the campus after that is easier still. Network effects in industrial markets work the same way they work in software. The first data center is hard. The thirtieth is routine.

That's the part of the story that gets missed when the conversation reduces to tax incentives. Tax policy didn't make Arizona competitive. Forty years of compounding industrial infrastructure did. The state's Computer Data Center program (ARS 41-1519) exempts qualifying equipment from transaction privilege and use tax for up to ten years, but it only kept the door open while the math worked itself out.

Sources