A data center can look deceptively simple from the road. A large rectangular building. Few windows. Security fencing. Cooling equipment. A substation nearby. Not much foot traffic.
Then the planning file arrives.
The building may need hundreds of megawatts of electricity over several phases. New transmission infrastructure may be required. Cooling may use significant water, or almost none, depending on technology and climate. Backup generators can create noise and air-quality questions. Battery systems introduce a different emergency-response profile. Blank façades and large compounds can reshape an employment district. A cluster of projects can consume industrial land faster than maps anticipated. The site may create relatively little daily traffic while placing extraordinary demand on utilities that are mostly invisible to ordinary zoning maps.
That is why data centers have become one of the clearest examples of a modern planning problem: a land use whose physical footprint, infrastructure footprint and economic footprint are very different sizes.
This article has one reader job: help planners, landowners and communities understand how to regulate data centers as a distinct land use without pretending they are ordinary warehouses, ordinary offices or power stations. It focuses on the planning job—classification, siting, compatibility, infrastructure coordination, evidence and long-term land-use control—without turning into an electricity-market or corporate-finance article.
1. Why the old “warehouse” analogy breaks
Data centers often occupy large, box-like buildings on industrial land, so early zoning codes sometimes placed them into warehouse, office or general industrial categories. That can be administratively convenient and analytically wrong.
A warehouse is primarily a storage-and-logistics use. Its land-use impacts are often shaped by trucks, workers, loading bays, operating hours and road access. A data center stores and processes information, but its critical operating systems are electrical supply, cooling, backup power, fiber connectivity and security. The most important off-site effects may occur through substations, transmission lines, generator testing, water demand and cumulative utility capacity rather than through daily vehicle trips.
The American Planning Association has argued for treating data centers as a distinct use precisely because targeted definitions let local governments regulate location and differentiate projects by floor area, site area, energy demand, water demand or other land-use characteristics. In 2025 and 2026, Loudoun County, Virginia—one of the world’s most data-center-intensive local jurisdictions—has been actively rewriting its approach to data center locations and use-specific standards.
2. The definition is the first planning decision
A zoning code cannot route an application correctly if it cannot identify the use.
A useful definition should distinguish a data center from ordinary server rooms inside another business, telecom switching facilities, research computing, cryptocurrency mining where treated separately, utility infrastructure and other digital facilities. The definition should identify the use by function and scale rather than brand name.
Possible defining characteristics include off-site data processing or storage, high-density computing equipment, dedicated cooling systems, enhanced electrical infrastructure, backup generation, battery systems and secure controlled-access compounds. The code may also need thresholds separating a small embedded facility from a purpose-built hyperscale campus.
For the deeper classification logic, see TPW-0166 — The Land-Use Definition.
3. Scale should not be measured by floor area alone
Two buildings of identical size can create radically different infrastructure demand.
A data center’s planning scale can be described through several variables at once:
- site area;
- gross floor area;
- ultimate electrical demand;
- number and size of generator units;
- cooling technology and water demand;
- substation and transmission requirements;
- battery or energy-storage capacity;
- phasing period;
- building height and rooftop equipment;
- distance to sensitive receptors.
The zoning threshold should follow the impact being managed. Floor area may work for façade or bulk controls. Megawatts may be more useful for grid coordination. Generator capacity may matter for air and noise review. Water intensity may matter in a constrained basin.
4. By-right or conditional? That is the governance hinge
One of the most consequential code choices is whether data centers are permitted by right, prohibited, or allowed only through conditional or discretionary review.
Loudoun County changed this balance in March 2025. Its Board of Supervisors adopted plan and zoning amendments that removed by-right treatment for data centers in several major industrial districts and made them conditional uses requiring special-exception approval. In 2026 the county is still working through a second phase focused on location policy and use-specific standards, including noise, height, onsite power generation and energy storage.
This is not proof that every city should use discretionary review. It is evidence that mature data-center markets can outgrow rules written when the use was less infrastructure-intensive.
For the general distinction, see TPW-0103 — The By-Right Approval and TPW-0104 — The Conditional Use Permit.
5. A conditional process should not become an unwritten moratorium
Discretion can improve site-specific review, but only if applicants know the findings they must satisfy.
A defensible data-center conditional-use system states the questions in advance: compatible location, utility feasibility, noise performance, visual treatment, water strategy, emergency access, cumulative infrastructure effects, generator testing, landscaping, decommissioning and compliance monitoring.
If the approval body can simply say “too much data center” without adopted criteria, the code replaces predictability with politics. The point of discretionary review is not unlimited bargaining. It is structured evidence-based judgment.
6. Location policy begins with the grid—but cannot end there
Data centers gravitate toward places with power, fiber, land and infrastructure. From an operator’s perspective, proximity to transmission and substations can be decisive.
From a town-planning perspective, that creates a feedback loop. Once a jurisdiction designates land for data centers, utility infrastructure may expand toward it. Once new utility capacity exists, more data-center proposals become feasible. The land-use map and energy map begin shaping each other.
The U.S. Department of Energy’s July 2026 draft National Transmission Needs Study identifies load growth from data centers, manufacturing and other large loads as a driver of additional transmission needs. That is a national-scale grid observation with a local planning consequence: a building permit can be part of a much larger infrastructure chain.
For the transmission-corridor owner, see TPW-0139 — The Transmission Corridor Map.
7. The substation is often part of the land-use project even when legally separate
A data-center campus may depend on a dedicated or nearby substation owned by the operator or utility. Planning documents should make this relationship visible.
Questions include land area, setbacks, transformer noise, visual screening, transmission connections, emergency access, oil containment where relevant, construction sequencing and whether utility land fragments the public realm.
Reviewing the main building while treating the substation as someone else’s future problem produces incomplete planning.
8. Power demand should be stated at ultimate build-out
Phased projects can make early demand look modest.
A campus may begin with one building and a fraction of its eventual electrical load. If the planning authority reviews only Phase 1, it can miss the infrastructure and land-use consequences of the approved envelope.
Applications should therefore disclose current, phase-specific and ultimate demand, together with realistic assumptions about when capacity is expected. The purpose is not to turn planners into grid engineers. It is to prevent the land-use decision from understating the project’s scale.
9. Water use is a technology question before it is a zoning number
Data centers do not have one universal water profile.
Cooling architecture, climate, equipment density and operational strategy can produce very different water demand. The Department of Energy notes in current 2026 guidance that data-center water use can vary widely and, in some cases, reach millions of gallons per day, while advanced cooling technologies can substantially reduce demand.
A planning code should therefore avoid assuming that every data center is automatically water intensive—or automatically water light. Require a project-specific water balance where water availability is material.
For the wider growth-and-water owner, see TPW-0089 — The Drought Capacity Map.
10. Ask for the operating water profile, not just annual consumption
Peak demand matters. Seasonal demand matters. Emergency demand matters.
A facility that uses more water on the hottest days may place stress on the system at exactly the moment other urban demand peaks. Reclaimed-water availability may differ by season. Drought restrictions may affect cooling strategies.
The planning question is therefore whether the project’s operating profile fits the capacity and resilience of the local water system—not merely whether an annual average can be supplied.
11. Noise has several sources and several clocks
A data center can generate sound from rooftop cooling equipment, chillers, fans, transformers, substations, generators, maintenance activity and construction.
Steady mechanical hum has a different character from episodic generator testing. Daytime construction noise has a different receptor profile from nighttime cooling plant. Low-frequency components can matter even when a simple overall decibel number looks acceptable.
That makes data centers a strong candidate for performance-based controls tied to receptors, measurement method, time period and operating mode. See TPW-0132 — The Performance Standard.
12. Generator testing is a planning event
Backup generators may rarely run for extended emergency periods, but testing can be regular and predictable.
A condition that focuses only on emergency operation can miss the everyday nuisance issue. Planning review should distinguish emergency use, routine testing, maintenance testing and commissioning. Testing hours, sequencing and acoustic controls can be designed accordingly.
Where air permits are separately required, those remain a different regulatory track. Planning should coordinate with them without pretending to replace them.
13. Onsite generation changes the land-use envelope
Some data-center proposals include significant onsite generation, whether for backup, resilience or other operational reasons. Once generation becomes physically substantial, the project can no longer be analysed as a building plus incidental equipment.
Fuel storage, emissions, stacks, noise, safety separation, maintenance access and visual mass may all change. Codes should identify the point at which onsite generation triggers additional review or a separate use classification.
14. Battery systems are not just cabinets beside the server hall
Uninterruptible power and grid-scale storage can involve substantial battery capacity. The land-use plan needs to coordinate fire access, separation, emergency planning, containment, ventilation and decommissioning with the applicable fire and building codes.
For the larger battery-siting owner, see TPW-0091 — The Battery Siting Map.
15. Building height is only one part of visual scale
Data centers can be visually dominant without being skyscrapers.
Very long façades, blank walls, security setbacks, rooftop mechanical equipment, substations and fenced compounds can create a scale unlike the ordinary employment buildings anticipated by an industrial plan.
Visual standards should therefore consider frontage length, articulation, equipment screening, roofline, landscaping, public-facing edges and the cumulative effect of multiple campuses—not just a maximum metre figure.
For the measurement owner, see TPW-0157 — The Building Height Rule.
16. Blank walls create an urban-design problem even in industrial districts
Industrial land is not design-free land.
A facility beside a highway interchange may be experienced mainly at speed. The same building beside a mixed-use edge, trail or ordinary street needs a different response. Landscaping cannot permanently hide a 300-metre blank façade.
Design requirements should follow context and avoid pretending a server hall must look like a retail street. The objective is legible, durable edges—not decorative theatre.
17. Security design can unintentionally privatise the street
Setbacks, fences, guardhouses and controlled access may be operationally necessary. If every data-center parcel uses deep defensive setbacks, however, an entire district can become a sequence of blank edges with little pedestrian legibility.
Planning can distinguish genuine security requirements from habits carried over from remote campuses. Fences can be integrated with landscape. Guardhouses can be placed without blocking public paths. Utility equipment can be screened without obstructing sightlines.
18. Buffers should respond to the receptor, not become a ritual number
A 100-metre buffer is not automatically enough, and a 300-metre buffer is not automatically necessary.
The right separation depends on source and receptor: transformer noise, generator location, building height, terrain, vegetation, existing roads, nearby housing, public space and other industry. Performance evidence can justify a smaller or larger physical separation.
This is another reason to combine dimensional standards with measurable outcomes rather than using setbacks as a substitute for analysis.
19. Industrial-land opportunity cost belongs in the plan
A data center may be an excellent use of one site and a poor use of another even if both are zoned industrial.
Some industrial districts are scarce, well-served locations needed for production, repair, distribution, urban services or businesses that depend on a large workforce. A very land-intensive digital facility can consume that capacity while creating a different employment pattern.
The planning question is not whether data centers create “enough jobs” in the abstract. It is whether this particular land is strategically intended for uses whose locational needs cannot be met elsewhere.
20. Do not confuse property-tax attraction with land-use suitability
Fiscal benefits can matter to local government, but the town-planning decision should still test land, infrastructure and compatibility.
A jurisdiction that zones purely for near-term revenue can lock scarce land and utility capacity into a use whose long-term infrastructure obligations were never fully modelled. Conversely, ignoring the fiscal consequences entirely can also produce unrealistic policy.
The correct division of labour is simple: finance evaluates public revenue and cost; planning evaluates whether the use belongs on the site and how it should operate there.
21. Cumulative impacts are where parcel-by-parcel zoning can fail
One data center may fit. Twenty may change the district.
Cumulative questions include total power demand, transmission corridors, substation proliferation, water demand, industrial-land consumption, construction traffic, generator testing, visual character and emergency-service capacity.
A local government that approves each parcel as if it were the first can lose control of the aggregate system. Comprehensive-plan policy should therefore define where clusters are intended, where limits matter and what infrastructure thresholds trigger a broader review.
22. The cluster map should include projects not yet operating
Operational facilities show today’s condition. Approved and queued projects show tomorrow’s.
A planning map should therefore distinguish existing, under construction, approved, pending and concept-stage projects where data is available. The same applies to substations and major transmission upgrades.
Otherwise a site may appear isolated on a current map even though several neighbouring projects are already entitled.
23. Fiber is necessary but usually not the planning bottleneck
Data centers need high-capacity connectivity, but fiber can often be routed with less visible land consumption than power infrastructure. The planning file should still identify major trench routes, rights-of-way and crossing constraints where relevant.
For the broader digital-infrastructure owner, see TPW-0088 — The Broadband Map.
24. Construction can be more disruptive than operation
Once operational, a data center can generate relatively modest daily vehicle activity compared with many warehouses or offices. Construction is different.
Large campuses involve earthworks, concrete, electrical equipment, transformers, cranes, heavy deliveries and long build-out periods. If multiple projects overlap, the district can experience years of construction effects.
Construction-management plans, haul routes, working hours, dust control and coordination with other major works may therefore deserve more attention than ordinary operational traffic assumptions suggest.
25. Emergency planning has to match the equipment stack
Emergency services need to understand batteries, fuel systems, generators, high-voltage infrastructure, secure access and cooling systems before an incident occurs.
The planning process can require an emergency-access and coordination framework without trying to write the fire code itself. Plans should show apparatus routes, gates, emergency contacts, shutoff points and areas that require specialised response.
26. Diesel storage and refuelling are land-use details with failure modes
Where backup systems depend on large fuel volumes, the site plan should consider tanker access, containment, spill response, separation and how refuelling occurs during an extended outage.
These systems may be heavily regulated outside zoning. Planning’s role is spatial: make sure the site can safely accommodate the operating pattern assumed by those other permits.
27. Heat rejection belongs in the physical model
Cooling equipment rejects heat. In dense clusters, planners should at least understand where major heat-rejection systems sit, how they interact with nearby public spaces or buildings, and whether the proposed configuration creates local microclimate or plume considerations that specialist analysis should test.
The planning authority does not need to become a mechanical engineer. It needs to know when specialist evidence is necessary.
28. Stormwater design can become unusually land hungry
Large roofs, extensive hardstanding, security roads and equipment yards can create substantial impervious area. A campus can therefore require major detention, infiltration or treatment infrastructure.
That land should be visible early in concept planning. Treating stormwater as an engineering afterthought can compress buffers and landscape later, creating a cycle of post-approval amendments.
29. Noise modelling should include future phases
A Phase 1 acoustic model can be compliant while the ultimate campus is not.
Modeling should identify the ultimate reasonable equipment scenario, including cumulative mechanical plant, transformers and testing regimes. Conditions can then tie each phase to the overall performance envelope.
30. Generator testing needs coordination across campuses
If five facilities each test generators independently, a neighbourhood can experience frequent noise even when each individual site complies.
Where data-center clusters are dense, local rules can consider cumulative scheduling, time windows and receptor-based performance rather than treating each compound as acoustically isolated.
31. Grandfathering rules need a date and an application test
When a jurisdiction changes data-center rules, projects already in the pipeline raise transition questions.
Loudoun County’s 2025 changes included a grandfathering resolution identifying how certain applications accepted before a stated date could continue under the earlier framework. That is a useful model of procedural clarity even if another jurisdiction chooses a different policy.
Transition rules should define the cutoff event: informal concept meeting, submitted application, complete application, accepted application, approval or permit issuance. Vague grandfathering creates disputes.
32. Pre-application review is especially valuable for data centers
The expensive problems are often utility and environmental problems, not architectural ones.
A serious pre-application process can surface power pathway, water strategy, substation location, noise receptors, transmission needs, emergency access, phasing and required technical studies before the applicant freezes a site plan.
See TPW-0180 — The Pre-Application Conference.
33. The completeness checklist should be use-specific
A generic industrial application checklist may not ask for ultimate megawatts, cooling-water demand, generator-testing schedule or substation phasing. Then the most important information arrives halfway through review.
Use-specific validation requirements can save time by asking for the decisive evidence on day one. See TPW-0176 — The Application Completeness Test.
34. A good data-center application separates facts from promises
“We intend to use efficient cooling” is not the same as a defined water budget. “We expect clean power” is not the same as an approved supply arrangement. “Generators will rarely run” is not the same as a testing schedule.
Planning decisions should be based on measurable assumptions that can become conditions where necessary. The more important the assumption is to the acceptability of the project, the less it should depend on marketing language.
35. Community benefits should not purchase compatibility
A project may fund infrastructure, public improvements or community programmes through lawful mechanisms. Those benefits do not erase a fundamental siting problem.
First ask whether the land use is acceptable with appropriate mitigation. Only then evaluate lawful public-benefit mechanisms. This keeps planning findings from becoming a price negotiation over impacts that should not be accepted in the first place.
36. Decommissioning deserves more attention than it usually gets
Digital infrastructure can become technologically obsolete faster than conventional buildings.
A data-center campus may still have valuable shell buildings, power connections and land. But some sites may become stranded if equipment, cooling systems or utility assumptions change dramatically.
Planning can require removal of disused external plant, safe handling of fuel and batteries, site stabilisation and a defined process if the use ceases. A decommissioning plan should focus on physical land-use obligations, not speculation about the company’s future business model.
37. Adaptive reuse should be tested before the code assumes demolition
Some data-center buildings may be hard to convert because of deep floorplates, security design, limited windows or highly specialised infrastructure. Others may have strong structural grids, generous floor loads and power connections valuable for new industrial or research uses.
Future adaptability can be improved by site planning that preserves street access, avoids unnecessarily fragmented parcels and keeps utility infrastructure organised.
38. The monitoring condition should collect planning data, not corporate secrets
Local governments may need enough operational data to verify conditions: water use, generator testing, noise complaints, completed phases or other site-specific metrics.
The reporting requirement should be proportionate. Planning does not need access to customer data, computing workloads or proprietary algorithms. Collect only what is necessary to monitor the land-use approval.
39. The “AI” label should not replace the land-use analysis
Artificial-intelligence demand is one current driver of data-center growth, but zoning should regulate the physical use rather than the fashionable workload inside it.
A building can shift from cloud computing to AI training without changing its planning impacts, or its power density can change dramatically while the corporate label stays the same. The code should focus on measurable site characteristics that remain relevant when technology cycles move on.
40. A worked example: the 80-megawatt industrial parcel
Imagine a 20-hectare industrial site proposed for a two-building data-center campus with an ultimate demand of 80 megawatts. A dedicated substation sits on the eastern boundary. Cooling is primarily air based, with limited water use. Forty backup generators are proposed. Housing lies 250 metres west beyond an arterial road.
A weak review asks whether “data center” is permitted in the industrial district and whether building setbacks comply.
A strong review also asks: Can the utility supply the ultimate demand? What transmission work is associated with the project? Where are transformers and generators relative to housing? What is the worst-case acoustic model at night and during testing? How are tests scheduled? What is the water balance? Does the substation fragment a planned street connection? What is the emergency plan for batteries and fuel? How do rooftop plant and blank façades appear from the arterial? What happens if three adjacent parcels convert to the same use?
That is the difference between checking a code and planning a place.
41. A worked example: the “small” expansion
An existing data center proposes a 15 per cent floor-area expansion. The building change looks modest. The new computing halls, however, double power density and require another substation yard plus 12 generators.
If the code measures materiality only by floor area, it may route the project as a minor amendment even though the infrastructure impact is major. The amendment system should therefore identify operational thresholds as well as dimensional ones.
42. A worked example: the water-constrained region
A proposed campus is technically permitted and close to transmission, but the region faces recurring drought. The applicant can choose between cooling systems with different water and energy profiles.
The planning condition should not dictate proprietary engineering without reason. It can, however, establish an approved water-demand envelope, require disclosure of cooling assumptions, secure alternative-water sources where part of the proposal and prevent later equipment changes from quietly exceeding the accepted resource profile.
43. A worked example: the industrial district becomes a data-center district by accident
Ten data centers are individually approved over six years in a general industrial area. Each complies with zoning. Manufacturers then struggle to find serviced sites because land values, power reservations and parcel assemblages have shifted around the new cluster.
No single approval caused the transformation. The absence of a cumulative land-use strategy did.
The remedy is not necessarily prohibition. It is explicit policy: identify where digital infrastructure is intended, where industrial diversity is protected and how much utility-intensive growth the district is meant to absorb.
44. The applicant checklist
- Confirm the use definition and approval route.
- State site area, floor area and ultimate build-out.
- Disclose phase-specific and ultimate electrical demand.
- Map substation and transmission dependencies.
- Provide cooling technology and water-demand assumptions.
- Model operational, transformer and generator-test noise.
- Show backup generation, fuel and battery systems.
- Provide emergency access and response coordination.
- Address building form, public edges and screening.
- Assess cumulative effects with nearby approved facilities.
- Define monitoring and decommissioning commitments.
- Keep technical assumptions measurable enough to condition.
45. The planning-authority checklist
- Define data centers as a distinct use where scale justifies it.
- Choose by-right, conditional or prohibited locations deliberately.
- Use impact thresholds that go beyond floor area where necessary.
- Coordinate land-use policy with utility planning without delegating zoning to the utility.
- Map existing, approved and pending clusters.
- Protect scarce industrial land where other uses have stronger locational need.
- Require receptor-based noise and environmental evidence.
- Write clear grandfathering rules when standards change.
- Monitor cumulative infrastructure effects.
- Keep fiscal attraction separate from basic compatibility findings.
46. The core idea
A data center is not difficult to plan because it is digital. It is difficult because the most important physical systems extend far beyond the server hall.
The building sits on one parcel. The electricity may depend on a regional grid. The cooling choice interacts with a watershed. The generators affect nearby receptors. The substation changes the site plan. The cumulative cluster can rewrite an industrial district. The data workload can change faster than the building.
Good town planning therefore treats data centers as infrastructure-intensive land uses whose approval must connect the parcel to the systems that make the parcel possible.
The planning question is not simply “Can a data center fit here?” It is “What else has to fit, function and remain resilient because the data center is here?”
Authoritative reference points
- American Planning Association — Managing AI Build-Out in a Winner-Take-Most World, including guidance on treating data centers as a distinct land use.
- Loudoun County, Virginia — Phase 2: Data Center Standards & Locations, current 2026 work on noise, height, onsite generation, storage and location policy.
- Loudoun County — Data Centers: Land Use Considerations, including the county’s 2025 shift from by-right to conditional review.
- U.S. Department of Energy — Harnessing Data Centers for Tribal Economic Development, current 2026 guidance on water, transmission, interconnection, fiber and community considerations.
- U.S. Department of Energy — 2026 Draft National Transmission Needs Study, identifying data-center load growth among drivers of new transmission need.