Main Street Will Not Get a Data Center. The Central Office Might
By Brian Newman, Contributing Editor
Pew Research reported in April that 67 percent of planned United States data centers are headed for rural areas, while 87 percent of existing facilities sit in urban ones. More than 1,500 facilities are in some stage of development, and 39 percent of them are planned for counties that have none today. Every distressed county in America read that number and asked the same question: why not here? The question is half right. The capital is coming. It is not coming to the square.
One caveat belongs on the table before any community plans against that figure. Pew classifies rural using the Census Bureau definition, which counts low-density areas on the outskirts of metropolitan counties. A parallel analysis by the Daily Yonder, using the narrower federal definition that excludes metropolitan counties entirely, finds a far smaller rural share. The direction of travel is not in dispute. The magnitude depends on which map is being read.
I recently drove through some rural towns in my part of North Carolina and was amazed at the number of empty downtown budlings. Unfortunately, a vacant 1920s storefront cannot host AI compute. The building has no substation adjacency, no structural floor loading, no chiller yard, and no truck access. Downtown revitalization and hyperscale compute solve different problems with different assets.
Site selection is not a beauty contest. It runs through four gates. Power carries the most weight, and it is now the binding constraint on the entire market. As it stands, median interconnection wait times are approaching five years, with roughly 2,300 to 2,600 gigawatts of capacity stalled in United States queues depending on the count. Fiber transport is second, and the relevant question is long-haul and middle-mile capacity rather than fiber to the home. Land comes third, flat and contiguous and buildable. Water comes fourth, with volume varying by cooling design. There is a fifth gate that rarely appears on any list and increasingly decides outcomes: speed to yes.
This is where telecom stops being a spectator. Copper retirement is emptying central offices in exactly the towns now appearing on developer maps. AT&T has described a central office built in the 1950s for 100,000 copper lines that now serves 3,800. Those buildings were engineered for continuous power, backup generation, cooling, physical security, and fiber termination. That specification sheet describes a small data center that already exists, already zoned, already sitting on transport. NVIDIA counted roughly 100,000 distributed network sites worldwide, spanning regional hubs, mobile switching offices, and central offices, with enough spare power to support more than 100 gigawatts of new AI capacity over time. Telefónica has begun converting roughly 100 retired copper exchanges into edge nodes of one to two megawatts each, with about a dozen already operational.
North Carolina illustrates the second row of the inventory. The state ranks among the leaders in data center construction starts this year. The viable sites are not on the retail block. They are the closed textile mills, tobacco processing plants, and furniture factories that carry heavy electrical service, water rights, rail, a municipal boundary, and in many cases a substation across the road.
Capture of that value is negotiated, and the record is uneven. The 2009 agreement that brought Apple to Maiden, North Carolina, committed the company to roughly one billion dollars over ten years in exchange for a 50 percent real property tax abatement and an 85 percent personal property tax abatement, against a floor of 50 full-time jobs. Investment at the site has since grown to roughly four billion dollars and employment to about 400. The capital arrived. The floor was set once, early, and low, and the town negotiated it before anyone understood what the asset would become.
The same structure is now playing out with sovereign hosts. Permitting on conventional land runs three to ten years, and tribal permitting authority can compress that dramatically, which is precisely why developers are knocking. Several Nations have declined. The Seminole Nation of Oklahoma passed a data center moratorium in March on a unanimous council vote, and the Sault Ste. Marie Tribe of Chippewa Indians adopted one in April. In higher education, strategist Ashley Northington argued in Tech Policy Press that institutions entering these partnerships as landlords risk a form of digital sharecropping. That essay is advocacy rather than neutral analysis, and the structural point survives the framing anyway. A land lease is the weakest available position, and it is the default.
The Argument in Four Parts
The compute will go where the power is. That question is settled. The open question is what the hosts walk away with, and the hosts include the operators. Telecom spent a decade treating retired central offices as a disposal program. The AI buildout has repriced them, and the industry has not adjusted its accounting.
Brian C. Newman is a telecom and AI strategy consultant, course creator, and former Verizon technology leader with more than 30 years of experience across wireless networks, 5G, network operations, infrastructure modernization, and emerging technologies. He helps organizations understand how AI, connectivity, edge computing, and digital infrastructure are reshaping business operations, real estate, public safety, and customer experience.
