Site selection and due diligence

The site decides much of a data center's cost, risk and performance before any engineering begins. Power availability determines when the campus can open and how large it can grow; climate determines how much energy and water the cooling system uses; fiber routes determine network latency and resilience; hazards such as floods, earthquakes and wildfires determine risk and construction cost; and local permitting, tax and workforce conditions determine schedule and operating cost. In recent years, power availability has become the dominant constraint: in many established markets, utilities quote years of waiting for new large connections, and site selection now often starts with the question "where can we get power, and when?" This lesson sets out the main site selection criteria, the data sources and studies used to evaluate them, a weighted scoring method with sensitivity testing, and the due diligence that follows a preferred site choice.

Selection criteria

• Power: available transmission or distribution capacity, the time and cost to connect, the utility's reliability record, the possibility of two independent supply paths, the energy price and tariff structure, and access to carbon-free electricity through the grid mix or power purchase agreements.
• Fiber and network: proximity to long-haul fiber routes, the number of independent carriers, diverse path options into the site, latency to major population centers and cloud regions, and cost of dark fiber or wavelengths.
• Land: area for the ultimate campus plus setbacks, substation, stormwater and future expansion; shape and topography; zoning; ownership and price; and road access for heavy equipment deliveries.
• Natural hazards: flooding (including future climate projections), seismic hazard, wildfire, severe wind and tornadoes, hail, ice storms, and geotechnical conditions such as soft soils or karst.
• Human-made hazards: proximity to airports and flight paths, rail lines carrying hazardous goods, pipelines, chemical plants, and other facilities that could cause fires, explosions or evacuations.
• Water: availability and cost of water for cooling (if evaporative cooling is used) and of sewer capacity for blowdown, plus regulatory and community attitudes to water use.
• Climate: dry-bulb and wet-bulb temperature profiles, which determine economizer hours, cooling energy and water use (Lesson 6).
• Workforce: availability of construction trades and of skilled operations staff within commuting distance.
• Incentives and costs: property tax, sales tax exemptions on equipment, economic development incentives, and construction cost levels.
• Community and approvals: zoning compatibility, community attitudes to data centers, noise sensitivity of neighbors, and the expected permitting timeline (Lesson 3).

Data sources and studies

Site evaluation draws on many sources. Utilities publish or provide on request information about substation and transmission capacity; for large loads, a formal load study or interconnection request gives binding answers. Flood maps, seismic hazard maps and wildfire risk maps come from government agencies. Fiber maps are available from carriers and commercial databases. Climate data comes from weather stations, with design values in the ASHRAE Handbook. Desktop studies are followed by site visits and, for shortlisted sites, by geotechnical investigations, environmental site assessments, surveys and title searches.

In the United States, sources include utility and regional transmission organization interconnection queues (for example, PJM, ERCOT or MISO), FEMA flood insurance rate maps, USGS seismic hazard maps and the ASCE 7 hazard tool for design values, and NOAA climate data.

Comparing the three candidate sites

The Aurora team has three candidates, summarized below. Scores range from 1 (poor) to 5 (excellent) and are based on the team's research.

Weighted scoring and sensitivity

A weighted score combines the criteria. Weights reflect the client's priorities; for Aurora, power carries the largest weight because of its impact on schedule and growth. Because weights are judgments, the team tests how sensitive the result is to them, varying each weight randomly within a plausible range and counting how often each site ranks first.

import numpy as np
# Weighted site scoring with a sensitivity test on the weights
criteria = ["power", "fiber", "land", "hazards", "water", "climate", "workforce", "incentives"]
weights = np.array([0.25, 0.12, 0.10, 0.12, 0.08, 0.13, 0.10, 0.10])
scores = {   # 1 (poor) to 5 (excellent)
    "Site A": [5, 3, 4, 4, 2, 5, 2, 4],
    "Site B": [3, 5, 3, 5, 4, 3, 5, 3],
    "Site C": [4, 4, 5, 2, 5, 4, 3, 5],
}
for site, s in scores.items():
    print(f"{site}: weighted score {np.dot(weights, s):.2f}")

rng = np.random.default_rng(2)
wins = {site: 0 for site in scores}
trials = 20_000
for _ in range(trials):
    w = weights * rng.uniform(0.7, 1.3, len(weights))   # each weight varies by up to 30 %
    w = w / w.sum()
    best = max(scores, key=lambda k: np.dot(w, scores[k]))
    wins[best] += 1
for site, n in wins.items():
    print(f"{site} ranks first in {n / trials:.0%} of weight scenarios")

Site A: weighted score 3.90
Site B: weighted score 3.76
Site C: weighted score 3.94
Site A ranks first in 32% of weight scenarios
Site B ranks first in 3% of weight scenarios
Site C ranks first in 65% of weight scenarios

The base scores are close: Site C leads narrowly with 3.94, followed by Site A at 3.90 and Site B at 3.76. The sensitivity test shows that Site C ranks first in about two-thirds of weight scenarios, Site A in about one-third, and Site B rarely. This tells the team that the choice between A and C depends on how the client values power timing, climate and workforce against land, water and incentives, and that Site C's main weakness, the partial floodplain, deserves close attention: if the flood risk can be engineered away by placing buildings on the higher part of the site, its hazard score improves and its lead becomes robust.

Fatal flaws and screening

Scoring compares sites that are all acceptable. Before scoring, sites must be screened for fatal flaws: conditions that rule a site out regardless of other strengths. Typical fatal flaws include no feasible power connection within the required schedule, zoning that prohibits data centers with no realistic path to rezoning, contamination requiring remediation beyond the budget, a buildable area too small for the minimum campus, or a regulatory moratorium on new large loads. Screening first saves effort and prevents a high score on other criteria from hiding a disqualifying problem.

Power deep dive

Because power is so critical, the team examines it in detail for the shortlisted sites:

• Capacity: how much load can the nearest substation and transmission lines carry, now and after planned upgrades?
• Timeline: how long will interconnection studies, utility approvals, substation construction and transmission upgrades take?
• Cost: who pays for the substation and upgrades, and what security deposits or contributions are required?
• Redundancy: can the site be served from two independent substations or transmission paths, or by a ring bus with multiple sources?
• Tariffs and contracts: energy rates, demand charges, minimum take-or-pay commitments for large loads, and any special tariffs for data centers.
• Carbon and on-site options: grid emissions intensity, availability of renewable energy contracts, and space for on-site generation or battery storage that could support grid services or bridge delays in utility capacity.

Many utilities now require large-load customers to sign agreements that commit them to minimum payments for a number of years, protecting other ratepayers if the data center does not materialize. These commitments affect project finance and should be identified early.

Due diligence on the preferred site

Once a preferred site is chosen, due diligence confirms that it can deliver what the scoring assumed before the client commits to purchase. Typical activities include:

• Title search, surveys and review of easements, rights of way and restrictions.
• Phase I environmental site assessment, and Phase II sampling if the Phase I identifies potential contamination.
• Geotechnical investigation with boreholes to determine soil bearing capacity, groundwater, and rock depth.
• Flood, drainage and stormwater studies, including future climate scenarios.
• Utility confirmation letters or agreements for power, water, sewer and gas.
• Zoning and permitting review with the municipality (Lesson 3).
• Traffic and access review for construction and operations.
• Ecological surveys for protected species and wetlands.
• Archaeological or heritage assessments where required.

Findings feed back into the design and cost estimate. A geotechnical report showing soft clay, for example, may require piled foundations that add significant cost and schedule.

Presenting the site recommendation

The site selection report presents the recommendation in a form the client can act on. It summarizes the brief's site requirements, the long list of candidates and the screening results, the scoring method and weights agreed with the client, the scores with supporting evidence, the sensitivity analysis, and the risks and mitigations for the preferred site. It also lists the due diligence still required and the decisions the client must make, such as authorizing a land option agreement or a utility load study deposit. Including the runner-up site, with the conditions under which it would become preferable, gives the client a fallback if due diligence uncovers a problem.

Teams sometimes fall in love with a site early and present the analysis as a justification rather than a comparison. Reviewers notice. A balanced report that shows the weaknesses of the preferred site alongside its strengths, and explains how they will be managed, is more persuasive and more useful.

The team should also record how the site decision could be revisited. If the utility later offers Site A an earlier connection date, or Site C's due diligence finds poor soils, the scoring model can be rerun quickly with updated inputs, and the client can see whether the recommendation changes.