A gigawatt-scale computing project changes the meaning of a good location before construction even begins. Cheap acreage can look attractive on a financial model, yet the value collapses if transmission capacity, interconnection timing, water availability, or reliable generation cannot support the intended load. Electricity demand from U.S. data centers reached about 176 TWh in 2023, while published projections place 2028 consumption between 325 TWh and 580 TWh, showing why large loads increasingly compete for infrastructure rather than simply for real estate.
That shift puts the substation, transmission corridor, generation fleet, and interconnection queue much closer to the center of site underwriting. A remote location can therefore become more valuable than an expensive metropolitan address when its electrical infrastructure can support expansion without forcing the project into a long transmission build. Fiber still matters, but it can travel farther than high-voltage infrastructure can be economically extended on the same schedule. Consequently, the strongest sites increasingly combine inherited industrial power infrastructure with enough land, water, labor access, and communications diversity to support a computing operation measured in hundreds of megawatts or more.
The Moment Fiber Stopped Picking The Site
For years, network proximity gave developers a practical reason to stay near established metropolitan corridors, where carrier hotels, enterprise customers, cloud connectivity, and dense fiber routes already existed, but the growing scale of AI facilities is making power availability a more decisive factor in site selection. That logic weakens when the electrical requirement approaches hundreds of megawatts because a network extension can follow a project while a major transmission upgrade requires engineering studies, equipment, rights-of-way, permitting, and coordinated utility work. Recent market analysis shows large AI-focused facilities moving farther from major cities as developers prioritize cheaper land and faster access to power.
That creates an unusual infrastructure equation in which a long fiber route can become acceptable when it unlocks substantially stronger electrical fundamentals. A long-distance communications build may involve trenching, conduit, splice points, route diversity, carrier agreements, and maintenance planning, but those costs can remain manageable relative to the consequences of waiting for major power infrastructure. The important distinction is that distance no longer carries the same penalty across every infrastructure layer. Fiber can be engineered around geography, whereas transmission capacity depends on physical network constraints that may require substantial system reinforcement before a large load can connect.
Why The Best Address For AI Is A Dead Factory
Retired industrial facilities deserve renewed attention because their infrastructure history can matter more than their current condition. Former steel mills, paper complexes, automotive plants, refineries, and other heavy industrial properties can offer useful infrastructure advantages because their locations may already have access to substantial electricity, transportation, water, and utility networks, although each site still requires separate technical and environmental evaluation. That does not make every former industrial site suitable, because contamination, demolition requirements, environmental restrictions, transmission constraints, and local permitting can materially affect whether an apparently attractive location can support a large computing development.
The more useful thesis is therefore not that abandoned factories automatically provide the answer, but that inherited infrastructure can reduce the amount of new infrastructure required when the site’s electrical, water, transportation, and utility assets align with the proposed development. Existing nuclear sites illustrate the same principle at a different scale, with analysis identifying dozens of operating or retired nuclear locations that could potentially host additional generation capacity. Nuclear-adjacent development also demonstrates why the energy asset itself can become a real-estate consideration when a large computing load requires firm electricity and lower dependence on long-distance transmission.
Water, Workforce, And Weather: The Three Filters After Power
Power availability only establishes whether the project can operate electrically; it does not establish whether the surrounding environment can sustain the project over decades. Water supply requires analysis of source reliability, permitted withdrawal, seasonal variability, competing demand, treatment requirements, and the cooling architecture selected for the computing load. Climate exposure adds another layer because heat, drought, flooding, severe storms, and other hazards can affect both utility reliability and the physical performance of infrastructure.
Workforce capacity creates an equally practical constraint because a large facility needs electricians, controls specialists, mechanical technicians, commissioning teams, security personnel, construction trades, and long-term operations staff. A rural site may offer inexpensive land while lacking sufficient housing, transport links, contractors, or technical labor within a practical operating radius, making workforce and supporting infrastructure important considerations alongside power and water availability. Climate resilience must also extend beyond the building envelope because substations, roads, fuel logistics, water systems, telecommunications routes, and regional generation assets can all influence the facility’s ability to maintain operations during disruptive events.
The Ghost Fiber Problem No One Put In The Site Model
A power-led site strategy can address a primary infrastructure constraint while creating a communications challenge when the selected location sits far from established network hubs. Long-haul fiber must deliver sufficient capacity, physical route diversity, carrier diversity, repair access, and predictable latency rather than simply reaching the property boundary. A single route that appears redundant on a commercial map may still share bridges, rail corridors, utility rights-of-way, conduits, or regional choke points with another route, leaving the facility exposed to one physical failure.
The financial model therefore needs to treat communications as infrastructure with its own failure modes instead of assuming that fiber can arrive after the electrical decision. Route surveys, diverse entry points, independent backhaul paths, splice protection, maintenance agreements, and realistic restoration assumptions can materially change the cost of a remote development. Moreover, the distance between the facility and major exchange points can affect latency-sensitive applications even when raw bandwidth remains abundant. The resulting TCO should capture construction, recurring transport charges, redundancy, maintenance, outage exposure, and the operational consequences of losing a primary route because these factors can become meaningful components of the economics of a remote development.
The Reactor Is The New Anchor Tenant
The defining question for a very large computing development is becoming less about how much land can be assembled and more about which energy system can support that land for the intended operating life. A nearby reactor can offer a powerful anchor where regulatory, commercial, cooling, and grid conditions align, while existing nuclear sites can also provide infrastructure advantages that make them relevant to large-load development. That does not make nuclear the universal answer, because every generation technology carries its own development, reliability, regulatory, environmental, and transmission constraints.
Site selection therefore needs an energy-first underwriting model that evaluates generation, transmission, interconnection, water, workforce, climate exposure, communications, and expansion as one connected system. A strong location may sit far from a major city if its available infrastructure can support electrical growth while engineered fiber preserves the connectivity required by the intended workloads. The weakest location may carry inexpensive land and excellent highway access yet fail because the grid cannot deliver the required capacity within the project’s commercial timeline. Ultimately, the decisive question becomes which energy anchor the project can access, how much infrastructure that anchor can support, and what additional investment remains before the computing load can operate reliably.
