Data Center Construction: Site Selection and Timeline

Data center construction starts with power, not land price. A parcel with cheap acreage and no realistic interconnection path is worthless, while a parcel with power nearby is worth paying for even at doubled land cost. This page is for enterprise, sovereign and operator teams screening parcels for hyperscale and AI builds. It orders the eight siting criteria by what disqualifies a site first, and it paces the construction timeline by permitting and long-lead equipment rather than the building itself.

TL;DR: Order the screens by disqualification: power, land, water, fiber, then the negotiable tier. Permitting alone can add 6 to 18 months on top of the 18 to 30 month construction window, per Anvilfield's buildout-phase analysis, and large transformers can run past three years. The schedule is set before site work begins.

What Actually Decides Whether a Data Center Site Works

Site selection begins by ruling sites out; scoring applies only to what survives. Teams that screen land price first can end up buying acreage they cannot energize and find out late in diligence. The eight criteria below are ordered so that the screens that can rule a site out outright come first.

The distinction matters for sequencing capital. A site that fails a disqualifying criterion fails at any price; a site that fails a negotiable criterion costs time or money. Budget option money only after the first four rows clear.

Power Availability and Interconnection: The First Data Center Site Screen

Power availability is the first disqualifying criterion in data center site selection because it is the one a buyer can least fix. Three factors decide it.

1. Capacity and date. The parcel's load must match the utility's deliverable MW within the program's energization date. There is no universal MW threshold, so confirm capacity and dates with the serving utility.
2. Source of the capacity. A nearby substation with headroom is a different program from a new-build interconnection. The interconnection queue sets the wait before a single volt reaches the parcel. Across US interconnection queues, the national median wait exceeds five years and PJM averages over eight, per LBNL's Queued Up data. That wait can be the long pole in the entire program.
3. Cost to reach the parcel. Grid extension and substation work feed into the all-in cost per megawatt of the build, so the further the grid must stretch to reach the site, the more that figure can move.

A parcel that clears all three is worth a premium on land price. A parcel that fails the first cannot serve the program at any discount.

Land: Acreage for the Campus and the Substation

Land is the second disqualifying criterion for a data center site, and the arithmetic is where budgets can go wrong. The parcel has to hold two separate line items: the campus footprint and the substation yard. Sizing only the data halls under-buys acreage.

Initial hyperscale campuses need roughly 100 to 200 acres, with the largest campuses exceeding 500 acres, per RSP Engineers' site-selection analysis. Plan the electrical infrastructure as its own line item, and confirm whether any campus figure you are quoting already includes it.

The substation figures come from RSP's substation area planning. Voltage levels, ownership models and transformer lead times are the engineering decisions behind those yard sizes, so settle substation planning and ownership models with the utility before fixing the acreage.

Price the parcel against the combined footprint of campus and substation yard. Large (50+ acre) US parcels relevant to data centers averaged about $244,000 per acre in 2024, with prices for 50+ acre parcels up 23% year over year, per TrueLook's construction cost breakdown. Acreage that looks cheap per parcel can be expensive per usable acre once the substation yard, stormwater retention, and setback requirements come out of the gross site.

Water: A Cooling-Design Dependency

Water is a disqualifying criterion for one design class and far less important for the other, so the cooling system decides whether the screen applies. A water-cooled hyperscale or AI facility can need millions of gallons per day, per RSP Engineers. Local supply may not be able to carry a load that size.

The fork:

• Air-cooled designs avoid that load, so the water screen matters much less for them.
• Water-cooled designs need either a substantial water utility commitment or on-site treatment and reuse, and the water question must be answered before the mechanical design is locked.

For AI workloads, higher rack densities push the economics toward liquid cooling, which is why the cooling design and the water screen have to be settled together, as the cooling equation explains. A parcel with constrained water may still work for a design that depends less on water, but the design and site decisions have to be made together.

Fiber: Route Diversity as a Site Screen

Fiber is a disqualifying data center site criterion, and the test is route diversity. A parcel served by one fiber path has a single point of failure for the whole campus: one cut or conduit failure can take every tenant offline. A strong site has multiple physically diverse routes, on separate rights-of-way, entering the parcel from different directions. This is our own screening judgment, reasoned from that single-point-of-failure logic.

What fails the test:

• A single conduit serving the parcel, regardless of the carrier count inside it.
• Multiple carriers sharing one right-of-way or one bridge crossing.
• Routes that are diverse on paper but converge on shared infrastructure before leaving the region.

We treat diversity as disqualifying because it depends on physical routes that carriers must build, and that is typically outside the buyer's control and schedule. Zoning and incentives can be negotiated; a fiber route usually cannot. Ask each carrier to document its physical route.

Permits, Disaster Risk, Workforce and Incentives: The Negotiable Tier

Zoning and permits, disaster risk, workforce and incentives cost time or money, so they are evaluated together, after the disqualifying criteria (power, land, fiber, and water for water-cooled designs) clear.

Zoning and permits. Zoning must admit generators, fuel storage and substation acreage for the design to proceed, per Autodesk's guide to data center construction. Check all three against the parcel's actual zoning. Permitting problems can often be worked through with time, which is why this row sits in the negotiable tier. But that time cost belongs in the schedule.

Disaster risk. Floodplain, seismic and severe-storm exposure can raise insurance cost and design requirements. The power redundancy topologies a design needs to survive regional events should be sized against the parcel's actual exposure.

Workforce. Construction trades and operations staff must exist within commuting range. Where local labor is thin, a project may need rotation or fly-in crews, which can add schedule and cost. Price that into the parcel comparison before mobilization.

Incentives. Tax abatements and utility rate agreements usually change a project's economics, which is why incentives are the last row.

The Data Center Construction Timeline

Permitting and long-lead equipment set the pace of a data center build. Construction typically takes 18 to 30 months once site work begins, but permitting can add 6 to 18 months in front of it, long-lead electrical equipment runs 12 to 18 months, and large transformers can exceed three years, per Anvilfield's buildout-phase guide. The critical path typically runs through the procurement and pre-construction phases.

Phases 2 and 3 can run in parallel, and both must finish before phases 5 and 6 can complete. A building with no energized electrical train cannot serve load. Schedule the program backward from the energization date: fix the equipment delivery dates first, then permit and construct toward them.

Facility long-lead equipment (generators, transformers, switchgear) is a separate procurement track from GPU and rack hardware, which carries its own lead times; the data center supply chain covers the hardware track. A program that manages only one of the two tracks can end up with a finished shell waiting on silicon, or racks waiting on a transformer.

How AI Workloads Change the Site and the Schedule

AI data center construction raises the stakes on site power, cooling and phasing. AI racks often draw 30 to 80 kW or more, versus roughly 25 kW for conventional air-cooled racks, per iRecruit's 2026 owner benchmarks. Denser racks concentrate more load in each building and leave less tolerance for a marginal power path.

Three consequences for site selection:

1. Power density can shift the binding constraint. With more compute per building, the MW requirement may limit the site before the land requirement does, and AI data center power consumption at the rack level sets that MW requirement.
2. Water and cooling design have to be settled together. Higher densities push the economics toward liquid cooling, and if the design depends on water, supply becomes a first-order disqualifier.
3. Phased energization deserves planning. If the first building is commissioned while later buildings are still being permitted or built, the site must support construction and operations traffic at the same time.

Because the critical path runs through procurement, ordering transformers and switchgear early, once site control is secured, can shorten the path to energization compared with waiting for the building permit to clear. The tradeoff is committing capital before permits are certain.

Density also shows up in cost. The same iRecruit benchmarks put 2026 US construction-only cost at $8 to 12M per MW for standard builds and $15 to 20M or more per MW for AI-optimized designs, and TrueLook's construction cost breakdown points the same way. Land, interconnection and design sit outside those figures, so total project cost runs higher; the cost per megawatt breakdown covers the full stack.

What Disqualifies a Data Center Site: The Checklist Before Option Money

Three facts decide whether option money is worth spending. Each corresponds to a disqualifying criterion (power, land, fiber, and water for water-cooled designs), and each can be checked before any capital commitment.

1. An interconnection path with a deliverable date inside the program schedule. Queue position and utility confirmation come first.
2. Acreage for the campus plus the substation yard. Run the combined footprint math (data halls, substation yard, stormwater, setbacks) before the price-per-acre comparison.
3. Route-diverse fiber and, for water-cooled designs, committed water. These depend on physical infrastructure around the parcel and are hard to fix after purchase.

A parcel that clears all three can proceed to the negotiable tier with a realistic view of schedule: permitting in front, long-lead procurement in parallel, and energization as the fixed point the whole program schedules against. Zettabyte's TITAN facility-design practice sequences site diligence and procurement around that energization date.

FAQ

How long does data center construction take?

Construction runs 18 to 30 months once site work begins, per Anvilfield. Permitting can add 6 to 18 months, and transformer lead times can exceed three years, so the full program is longer than the build.

How many acres does a data center need?

An initial hyperscale campus needs roughly 100 to 200 acres, per RSP Engineers. A substation yard needs 2 to 4 acres at 100 MW and 10 to 15+ acres at 500+ MW; plan it as a separate line item.

What is the first step in site selection?

Confirm the interconnection path: available capacity and a deliverable date inside the program schedule. Every other criterion, including land price, is evaluated after that.