Edge Data Center Site Selection: Location, Power and Zoning

An edge data center is a small facility placed close to the people, devices or networks it serves. There is no single size threshold. What defines it is purpose: it moves computing closer to the user than a distant cloud region or hyperscale campus would. That purpose drives every other site decision, from parcel size to the type of utility service required.

In practice, the edge comes in layers. Cloud providers run metro-level edge zones that act as extensions of a parent cloud region; AWS describes its Local Zones as placing compute, storage and database services close to population centers, and its Wavelength Zones as embedding them inside carriers’ 5G networks.¹⁰ Below that sit telecom aggregation points and central offices, then tower-base modules, then equipment on customer premises.

This guide covers the land and site questions for the smaller layers. Our guide to site needs by data center type compares edge with colocation and hyperscale profiles side by side.

Distance sets a floor on delay. Light in fiber travels about 200,000 km per second, so every 1,000 km of fiber adds roughly 5 ms each way before any routers or switches are counted.¹ Fiber routes are rarely straight, so the real path is usually longer than the map distance.

That floor is why edge sites exist. When AWS introduced Local Zones, its Los Angeles example sent traffic to a zone in the same metro instead of to the full region in Oregon, cutting latency to single-digit milliseconds.⁴ AWS lists machine learning inference, real-time gaming, live video and media production among the target uses for Local Zones, and AR/VR and connected vehicles among those for Wavelength.¹⁰

AI is adding a new reason for edge capacity. JLL’s 2026 outlook expects inference demand to require geographic distribution to reduce latency, driving regional deployments and systems at the edge.¹² Our guide to AI training vs. inference site needs explains why inference, unlike training, is pulled toward users, and latency requirements by workload sets out the tolerances by application.

The result is a different screening order. For a hyperscale campus, the first question is how many megawatts can arrive and when. For an edge site, the first question is which users, carriers and cloud on-ramps the site can reach within the latency budget. Acreage is rarely the constraint; a few acres, or a corner of an existing compound, is often enough.

Tower compounds

Cell tower sites are already networked, already leased and spread across every metro. American Tower began deploying small modular edge data centers at tower sites in several U.S. cities in 2020.¹³ Purpose-built modules for this setting are compact: Vapor IO’s VEM-150 Vapor Edge Module supports 150 kW of IT equipment in six racks inside a unit about 30 feet long, and the company has since replaced it with a larger VEM-180 for new deployments.²

Central offices

Telephone central offices are another inventory of networked buildings in town centers. Verizon has described retrofitting central offices, where copper networking is being retired, into data centers for edge inference, noting that the buildings are already power-ready and permitted.⁵ STL Partners projects telco-operated AI factories and network edge sites at more than 3.5 GW of IT capacity globally by 2030, with far more of it in larger AI factories than at the network edge.¹¹

Metro edge buildings

The largest edge sites look like small colocation facilities. American Tower’s Raleigh edge data center was planned at about 4,000 square feet with 1 MW of initial power and up to 4 MW at full capacity, and the company said it had identified more than 1,000 of its sites able to handle multiple megawatts.³ Its broader data center business reported 30 operating facilities across eleven U.S. markets at the end of 2025.¹⁴

Edge loads are small by data center standards, which changes the utility conversation. A site of a few hundred kilowatts to a few megawatts is typically served from the local distribution system rather than a dedicated transmission substation. That avoids the multi-year studies of large-load interconnection, but it makes the capacity of the nearest feeder and distribution transformer the deciding factor. Confirm available capacity, voltage and phase with the utility early.

AI hardware is raising the stakes. A single liquid-cooled GPU rack can now draw about 132 kW, close to the total IT capacity of an early edge module.¹⁵² An edge site planned for conventional servers may not have the power or heat rejection for even a few modern inference racks. See high-density GPU racks and site power.

Edge sites still need backup power, and most use diesel or natural gas generators. Under EPA’s engine rules, emergency engines may run for maintenance, testing and limited non-emergency purposes for up to 100 hours a year, with emergency operation itself unlimited.⁹ Many states and air districts add their own permits and limits, and generator noise is often the issue neighbors raise first. Our guides to backup generators and fuel sizing and air permits for data center generators cover the details.

Zoning is often the least predictable part of an edge project, because many codes were not written with small data centers in mind. In 2025, Phoenix planning staff reported that data centers were not defined or listed in the city’s zoning ordinance and proposed a text amendment to add a definition along with location and design standards.⁷ Where the use is undefined, staff may treat an edge site as a utility, a warehouse, an office accessory use or something requiring a special permit.

Newer ordinances are starting to separate small facilities from large ones. A 2025 DeKalb County, Georgia, draft text amendment proposed minor, major and campus tiers based on floor area, electrical load and whether a substation is needed, with minor facilities allowed as an accessory use in the office-institutional district.⁸ Drafts do not always survive: the DeKalb Board of Commissioners voted to deny the amendment in June 2026.¹⁶ A Paris, Texas, ordinance names “edge data center” as a use in its central area district and limits street-level frontage for it.¹⁷ Our guide to data center zoning ordinances and overlay districts tracks this trend.

Tower sites raise a specific question. Section 6409(a) of the federal Spectrum Act streamlines “eligible facilities requests,” which are modifications to existing towers or base stations that involve the collocation, removal or replacement of transmission equipment without substantially changing the structure; for towers outside public rights-of-way, excavation more than 30 feet beyond the current site is one trigger for a substantial change.⁶ A data center module is compute equipment, not transmission equipment, so do not assume it rides on that process. Expect ordinary zoning and building review, and confirm with the jurisdiction and counsel.

An edge screen is short but specific. The aim is to confirm that a modest amount of power, strong fiber and an approvable use can come together quickly on a small footprint.

1. 01Define the latency target and the users, carriers or cloud on-ramps the site must reach, then check the fiber path, not just the distance. See fiber connectivity for data center sites.
2. 02Ask the utility what the nearest feeder can deliver now and after upgrades, and how long upgrades take.
3. 03Read the zoning code’s definitions and use tables for data centers, utilities and accessory uses, and look for pending amendments.
4. 04Plan backup power, fuel storage and generator testing around neighbors and air rules.
5. 05Leave room for denser hardware later: AI inference racks can need several times the power and cooling of conventional servers.

Edge parcels are judged differently from large campuses, and many sites that are too small or too constrained for hyperscale are good edge candidates. If you have one in mind, we can get a site reviewed against these criteria.