Data Center Campus Development: Site Criteria, Power Strategy, and AI Workflow

Campus-scale data centers are power, land, fiber, water and permitting problems before they are real estate projects.

Data center campus development means planning multiple data center buildings on one coordinated site, usually delivered in phases as power, tenants, capital and permits become available. A campus might start with 48 MW or 96 MW of critical load, then expand toward several hundred megawatts if the utility, interconnection position, land envelope and community approvals can support it.

The mistake is treating the campus as a larger version of a single facility. It is not. Campus development is closer to infrastructure development. The first question is not, 'Can the building fit?' The first question is, 'Can the site be energized, cooled, connected and entitled on a schedule that matches customer demand?'

JLL's 2026 Global Data Center Market Outlook estimates nearly 100 GW of new data center capacity could be added globally between 2026 and 2030, roughly doubling global capacity. That growth will not be distributed evenly. It will concentrate where developers can prove power, land control and execution certainty earlier than competitors.

What makes a data center campus different?

A single data center is usually underwritten around one building, one energization path and one primary delivery schedule. A campus is underwritten around option value. The developer is buying the right to deliver capacity over time.

That changes the diligence stack. The core questions are:

• How many total megawatts can the site realistically support?

• Which megawatts are committed, studied, reserved or speculative?

• Can substations, transmission upgrades and backup generation be phased?

• Does the land plan preserve expansion paths for buildings, yards, roads and stormwater?

• Can fiber diversity be delivered from more than one carrier route?

• Can water, wastewater, air permits, noise limits and local politics survive full build-out?

The answer is rarely binary. A site may support 96 MW quickly, 192 MW after a substation upgrade and 300 MW only if a transmission project lands on time. That distinction matters more than headline acreage.

Power is the gating variable

Berkeley Lab's 2024 United States Data Center Energy Usage Report, prepared with the U.S. Department of Energy, estimated U.S. data center electricity use at 176 TWh in 2023 and projected a range of 325 TWh to 580 TWh by 2028. EPRI's 2026 Powering Intelligence work goes further, projecting U.S. data centers could consume 9% to 17% of national electricity by 2030 under higher-growth AI scenarios.

For campus developers, this means power diligence has to move from a checklist item to the first underwriting gate. A serious campus plan should separate four categories:

1. Existing service capacity that can be delivered under current infrastructure.

2. Utility-confirmed expansion capacity with defined scope, cost and schedule.

3. Interconnection-dependent capacity tied to studies, network upgrades or queue position.

4. Developer-controlled supplemental power, including temporary generation or behind-the-meter assets.

Only the first two categories should be treated as bankable in a near-term development schedule. The third can support option value. The fourth can improve resilience or bridge timing but brings emissions, permitting, fuel and operating complexity.

AI can help here by tracking utility filings, queue updates, substation procurement, tariff changes and load forecasts across markets. It cannot turn a speculative interconnection study into delivered capacity.

Land has to work at full build-out

Campus sites need more than large parcels. They need land that can absorb electrical infrastructure, security setbacks, construction laydown, stormwater, future buildings and internal circulation without creating conflicts in phase three.

The practical screen starts with acreage, zoning, topography and environmental constraints. Then it tightens around buildable area, floodplain, wetlands, access, easements, adjacent uses and noise receptors. Industrial zoning helps, but it is not enough if neighbors, school districts or local officials see the campus as a utility burden with limited employment density.

Developers should model the site as a sequence, not a diagram. Phase one has to work while phase two is under construction. Substation yards cannot block later buildings. Temporary construction access cannot become the only route for operations. Stormwater assumptions need to hold after full impervious coverage, not just first delivery.

Fiber and latency still matter

Power decides whether the campus is viable. Connectivity decides which customers it can serve. Hyperscale AI training, cloud availability zones, enterprise colocation and edge workloads have different requirements for carrier diversity, latency, dark fiber availability and route redundancy.

A campus diligence package should map known long-haul routes, metro fiber, carrier points of presence, railroad or highway rights of way and practical trenching constraints. It should also identify single points of failure. Two carriers on the same physical corridor are not true diversity.

The AI workflow is useful because fiber evidence sits in messy places: carrier maps, easement records, construction permits, route announcements and local utility coordination documents. Agents can collect and reconcile those records faster than an analyst team. Network engineers still need to validate the design.

The AI workflow for campus diligence

A good AI-assisted campus workflow has five steps:

1. Define the target capacity, customer profile, timing and minimum power threshold.

2. Screen parcels against power, zoning, fiber, water, access, environmental and adjacency constraints.

3. Build a phased capacity model that separates confirmed, probable and speculative megawatts.

4. Generate issue logs for entitlement, utility, water, noise, air permitting and community risk.

5. Produce an investment committee memo with evidence links, open questions and human review flags.

The human judgment layer sits at the hard edges: utility credibility, community politics, customer fit, capital timing and technical design. AI should shorten the search and make the assumptions visible. It should not bless the site.

Campus development rewards teams that can say no early. If power is uncertain, fiber is fragile, water is politically exposed or phasing requires heroic assumptions, the answer is not more modeling. The answer is to move on before the option premium becomes sunk cost.