How Should Cities Plan for AI Data Center Power in 2026?

Direct Answer for Urban Planners

Cities should plan for AI data center power as a metropolitan infrastructure problem, not merely as a utility application or a private building project. An AI facility consumes electricity for servers, cooling, electrical conversion, storage, controls, and increasingly dense accelerator racks. Its demand may arrive in a concentrated 50- to 200-megawatt block, while the substation, transmission lines, generators, water lines, roads, and public approvals needed to support it can take several years to permit and construct. Urban planning should therefore connect land-use decisions to utility capital plans, realistic buildout schedules, emergency planning, climate policy, and measurable community benefits. The central question is not simply whether a data center is legal at a proposed parcel, but whether the city can serve its electrical, water, transportation, and emergency-service requirements without transferring unreasonable costs or risks to existing neighborhoods. As of September 29, 2026, this matters because proposed projects in the United States can be comparable in scale to major industrial customers, while some planned campuses are measured in multiple gigawatts.

Also worth reading: How Much Data Center Substation Capacity Does a 100 MW AI Campus Need in 2026? · How Should Cities Scale Urban Digital Twin Infrastructure Without Creating Expensive Data Silos? · How Long Will a Data Center Grid Interconnection Take in 2026, and What Can Developers Do?

A sound plan separates four different issues that are often incorrectly combined. The first is the facility’s nameplate electrical demand. The second is its expected load at full operation, which may be lower initially. The third is the utility infrastructure required to serve that load, including substations and transmission upgrades. The fourth is the physical and fiscal footprint of the campus, including roads, drainage, fire protection, water, land remediation, tax agreements, and decommissioning obligations. AI data centers differ from conventional offices because each unit of occupied space can require vastly more power and cooling. A proposal claiming only that it will create temporary construction jobs or increase the tax base is incomplete. Municipalities need both an infrastructure-capacity analysis and a public-interest evaluation before granting entitlements.

From Grid Request to Land-Use Approval

The power connection process usually begins with a preliminary utility study based on the developer’s equipment list, expected utilization, redundancy, and expansion phases. Planners should ask for a minimum electrical load, a probable operating load, a maximum coincident demand, and the expected commissioning date for each phase. They should also request the location and voltage of the nearest substation, the proposed transmission or distribution construction, the utility’s upgrade estimate, and any nonstandard equipment. It is important to distinguish customer-funded network upgrades from upgrades paid by the broader rate base, because those costs affect residential bills and political acceptance. A data center should not receive a tentative rezoning or site plan before its utility studies identify dependencies such as a second substation, a new switching station, or a transmission line crossing another jurisdiction.

A planning department should require evidence that generation and utility capacity will be available when the campus is scheduled to operate. A letter merely stating that the site is within a service area is not enough, because an existing wire crossing a property boundary does not guarantee spare capacity. The relevant questions include how many megawatts are currently available, how much firm capacity is expected, which project has queue priority, and whether the utility can offer an uninterrupted service plan without lowering reliability elsewhere. Cities should examine whether backup generation is permitted under air-quality rules and whether fuel storage, exhaust, heat rejection, or emergency notifications affect neighboring land uses. Projects reported as being 7.65 gigawatts should be treated as potential campus pipelines unless a permit or interconnection agreement establishes a smaller committed first phase. Planners must analyze phases rather than accepting a developer’s ultimate headline capacity as an immediate load.

Why AI Demand Changes the Urban Planning Test

AI workloads are unusually electricity intensive because training and inference require dense processor or accelerator configurations. Data centers also operate continuously, so a large facility can create a sustained baseline demand even when computing activity varies. Some modern accelerators are designed for high-power rack densities, making traditional air-cooling assumptions unreliable. The old practice of calculating mechanical cooling from floor area can therefore understate the required electrical service and cooling plant. Municipal reviewers should ask whether the project uses direct-to-chip liquid cooling, rear-door heat exchangers, immersion systems, or hybrid arrangements. They should require updated one-line diagrams and heat-management plans with every major equipment revision because a server-generation change can materially alter both electrical demand and water use.

Demand growth is also uncertain. Efficiency gains per computation may lower energy use, yet more computations, larger models, and greater deployment can increase total consumption. That distinction prevents two opposite planning errors: assuming that every new AI service will multiply electricity demand without limit, or assuming that rapid efficiency improvements make capacity planning unnecessary. Government, businesses, and utilities should work with interval load forecasts, rather than a single annual average, because a campus constrained to a limited number of substation feeders may be more valuable to the grid than a larger but interruptible industrial load. A developer might also agree to staged service, demand-response participation, on-site generation, or curtailment during grid emergencies. Such commitments should be enforceable through the planning approval, utility agreement, or operating permit rather than left in a nonbinding memorandum.

Infrastructure, Siting, and Community Effects

Power availability does not make a site suitable. Data centers can require large parcels, secure utility corridors, redundant fiber routes, stormwater systems, water connections, road improvements, and heavy equipment deliveries. Cities should preserve routes for future transmission lines and avoid approving industrial subdivisions that obstruct a planned corridor. A site plan should distinguish indoor compute areas from warehouses, office space, parking, generation equipment, cooling plant, and future expansion areas. It should show setbacks, visual effects, noise sources, fuel storage, emergency access, and the location of exhaust or heat-rejection equipment. Local zoning may classify the use as data processing or light industry, but planners should consider adopting project-specific conditions for unusually large electrical loads instead of relying on generic use classifications.

Community effects extend beyond electricity rates. A large data center can add road wear, noise, water demand, fire risk, and pressure to the local tax base. It may also produce little permanent employment because the facilities are highly automated, while construction employment can be substantial but temporary. Municipalities should reject unsupported claims that a project will create thousands of permanent jobs. Better practice is to require an independently reviewed employment estimate, distinguish construction from operation, state the wage assumptions, and identify the local service providers involved. Tax-increment financing, special assessments, utility fees, and property-tax abatements should be disclosed in present-dollar and lifecycle terms. Benefits payments may help fund road, fire, or utility work, but they should not be presented as evidence that the project is automatically beneficial to the host community.

The first practical step is to create an AI data center project team that includes planning, building, fire, public works, water, environmental health, emergency management, the utility, and the school or tax authority affected by the proposal. The team should establish a standard pre-application package and a single schedule for technical review. That package should contain the parcel’s existing and proposed zoning, acreage, floor area, computed equipment load, power-usage effectiveness targets, water balance, backup-power details, utility letters, construction schedule, tax proposal, and community-benefit plan. Review should occur before a rezoning hearing wherever possible. A public debate over benefits is much less useful when the technical basis of the project remains incomplete.

The city should also conduct scenario testing. A low case could represent an initial 20- to 50-megawatt phase, a central case could include the project’s first several buildings, and a high case could test the developer’s stated pipeline. Each scenario should identify annual electricity demand, peak demand, backup duration, substation loading, water consumption, road traffic, emergency staffing, and projected property or utility revenue. Sensitivity analysis should change the computer configuration, utilization rate, cooling method, construction delay, and on-site generation level. Reviewers should ask what happens if the project opens at only 30%, 60%, or 100% of its first-phase design. This approach exposes hidden commitments, such as infrastructure built for a distant second phase or tax incentives granted before sufficient construction occurs.

City documents should define enforceable checkpoints. Before site-plan approval, the applicant may need an initial utility study; before building permits, a detailed service plan; and before certificate of occupancy, evidence that the required substation and generation work is complete. A planning commission can require quarterly progress reports during construction and annual operating reports after launch. Reports should show actual peak and average load, electricity sources, backup-power testing, water consumption, air emissions, jobs, tax receipts, community payments, and corrective actions. A developer should post financial security for roads, drainage, or other public work and fund decommissioning. The approval should not assume that a company’s current market value guarantees future performance, so closure planning should identify who pays for shutdown, equipment removal, contaminated land, and restoration of utility infrastructure.

Alternatives, Trade-Offs, and Power Strategies

There is no single replacement for an AI data center connection. A smaller phased facility can reduce initial infrastructure exposure but may still require the same ultimate substation investment. Colocation consolidates many customers in one facility and can improve utilization, but its operator still controls a large concentrated load. Edge computing places some inference capacity near users and can reduce network latency, although it may distribute power demand across numerous smaller sites rather than eliminate it. Efficiency improvements reduce energy per task, but total demand can continue rising as usage expands. Renewable-energy contracts can address emissions associated with consumption, but new generation, transmission, and firm capacity may be needed before the contract has physical effect. The city should evaluate these options on their own terms instead of treating any one measure as a complete solution.

On-site solar, batteries, fuel cells, reciprocating engines, and combined-cycle generation can improve resilience, but they create different safety, air-quality, and land-use effects. A battery system can shift some consumption and provide short-duration backup, yet it does not reproduce a full generation fleet for extended periods. Gas generation may offer dispatchable capacity while fuel systems and emissions controls are developed, but it can expose communities to price volatility and local pollutants. Nuclear, hydro, geothermal, wind, and other sources may be part of a broader resource mix, but planners should resist claims that a proposed facility receives carbon-free power merely because a corporate purchase or future annual matching claim exists. Hourly matching and reliable additional capacity are more demanding than annual accounting. A balanced power plan may combine efficiency, flexible operation, on-site storage, utility investment, long-duration contracts, and transparent emissions reporting.

The urban alternative is sometimes not approving the project or accepting a smaller version. Municipalities could reserve industrial land for other employers, protect corridors needed for future grid expansion, or use performance zoning so the same electricity demand produces more jobs and less public cost. They can also encourage renovations of existing industrial sites where grid and transportation access are already adequate. However, moving a data center from one municipality does not automatically benefit the region; it may merely transfer local impacts and consume another community’s capacity. Regional planning should therefore occur across utility service territories and adjoining jurisdictions. The best alternative is the option with the best combination of deliverable power, lower social cost, reliable infrastructure, compatible land use, and enforceable local benefit, not automatically the project with the highest stated investment amount.

Common Mistakes and Cost Questions

One common mistake is using the nameplate rating as the expected demand and treating every announced watt as guaranteed load. Another is accepting a utility “will serve” letter without a construction date, cost allocation, or identified network upgrades. Cities can also make the error of dividing a campus capacity claim by the number of buildings and assuming that every phase will operate independently. It is a mistake to compare the developer’s property-tax promise with only the value of a new substation, while ignoring customer-funded lines, system-reserve charges, transmission upgrades, or public safety services. Reviews should separate direct project costs from system costs and indirect costs borne by the city, residents, and other customers.

No dependable universal price should be quoted for serving an AI data center. Site-specific costs can differ by tens or hundreds of millions of dollars because they depend on distance to substations, voltage level, network congestion, local generation, transmission access, redundancy, water infrastructure, and required environmental controls. A 100-megawatt customer connection is not a standardized commodity, and a total project budget is not the same as a utility connection fee. Municipal staff should require an engineer’s estimate followed by binding bids or negotiated schedules. They should also document who pays each upgrade, whether costs enter the utility rate base, what security is required, and what happens if a later tenant is smaller or larger than the original model. Publicly stating “millions in investment” without a cash-flow breakdown is not cost control.

Planners should likewise scrutinize water and energy claims. Liquid cooling can reduce dependence on evaporative cooling, but it is not automatically waterless. Evaporative systems are most water intensive in hot, dry weather, while once-through or hybrid systems can have different local effects. A project should provide a monthly and peak-season water balance, including make-up water, discharge, water rights where applicable, and drought-response measures. Electricity-use-effectiveness targets should be accompanied by a defined measurement boundary. A city should not accept a sustainability rating based solely on purchased renewable-energy certificates while ignoring on-site fuel generation or grid constraints. Operational reporting is the practical test: a project committed to efficient design should be able to show measured performance after it begins operating.

When Cities Should Act—and What to Do Next

Cities should act before the first major application becomes difficult to change. For markets already receiving multiple data-center proposals, a 12- to 18-month policy-development process can include inventory, consultant selection, stakeholder meetings, and draft standards. A jurisdiction facing its first proposal may use a conditional permit while it builds the necessary data, but it should not grant full final approval solely to avoid appearing anti-development. The immediate response should be a written information request, a joint utility-planning meeting, and a public map of current substations, transmission constraints, planned upgrades, industrial districts, and candidate corridors. Those actions improve factual understanding without deciding the merits before the evidence is available.

Public participation is especially important because utility costs and land-use effects extend beyond the host parcel. Planning commissions should hold at least one technical work session and one broader public hearing, with materials available far enough in advance for residents and businesses to examine them. The city should publish assumptions, conflicts of interest, contracts, and changes between draft and final proposals. A community advisory group may be useful, but it should not replace elected decision-making or technical review. Residents should receive plain-language comparisons of the project’s first phase and ultimate plan, the effect on projected rates, emergency response, water, taxes, jobs, and alternatives. The process should not treat objections as proof of bad faith; neither should it treat industry support or tax revenue as proof of public benefit.

By September 29, 2026, the defensible urban-planning position is that AI data centers can support economic activity and digital services, but their scale can exceed ordinary development assumptions. Cities should welcome appropriately sited projects only when power, land, emergency services, finances, and community commitments are demonstrated. They should connect every major entitlement to phased service, verified infrastructure, monitoring, and restoration duties. The strongest policy is neither an automatic ban nor unconditional approval, but a process that assigns costs transparently and measures actual results. Over time, reported electricity use, outages, bills, jobs, tax receipts, water consumption, and emissions will provide better evidence than promotional forecasts. A city that follows that approach can make investment decisions that are economically credible, environmentally accountable, and resilient to the uncertain pace of AI development.