The Site Check: How to Verify a Data Center Site Before You Spend on Diligence

The framework in this article is AIDCKOL's own working method, published so that landowners, developers and operators can judge and use it. It is not an industry standard, and it is not a substitute for formal diligence.

Every data center project begins with the same economic problem. A full diligence programme — utility load studies, title work, environmental assessments, geotechnical investigation, detailed permitting analysis — is thorough, slow, and expensive. Run it on every candidate site and the budget is gone before a viable project is found. Run it on too few, and a fatal flaw surfaces late, after the money is spent.

The practical answer is sequencing: a disciplined, evidence-based initial screen that answers the questions most likely to kill a project, at a fraction of full diligence cost, before the expensive work begins. That screen is what we call a Site Check.

Step zero: define the requirement before judging the site

A site cannot be screened in the abstract. "Is this a good data center site?" has no answer until the requirement is fixed: target capacity in MW, how it phases, the date it is needed, the redundancy the design assumes, and the cooling strategy — because cooling strategy determines how much water the project needs and how much power the cooling itself consumes. A 10 MW edge facility and a 200 MW AI campus are different projects sharing a name; a site that fails one may be adequate for the other. Every Site Check starts from a written requirement, and every conclusion is relative to it.

The eight screening areas

A Site Check examines eight areas, in an order driven by how often each one ends projects:

1. Land and title basics. Parcel identity (APN), ownership as shown in public records, parcel size and configuration, existing easements visible in public sources, and obvious encumbrances. This is a public-records review only — it identifies questions for a title commitment to answer later; it does not replace one.

2. Utility service territory. Which utility actually serves the parcel, and which transmission operator's infrastructure is nearby. Service territory boundaries decide who the counterparty is for everything that follows, and they are frequently not who a seller assumes. This is a factual lookup, and getting it wrong invalidates the rest of the power analysis.

3. Power feasibility and evidence. The heart of the screen. What capacity exists at or near the site today, what has been requested, what has been studied, and what — if anything — has been committed in writing? Where are the nearest substations and transmission lines, and what does the utility's published material (integrated resource plans, load forecasts, large-load process documents) say about timelines in that area? The next section sets out how we classify the answers.

4. Zoning and permits. The current zoning designation, whether data center use is permitted by right, conditional, or prohibited, the jurisdiction and process that would decide a change, and any local rules already adopted specifically for data centers. The question is not "can this ever be approved" but "what is the documented path, and what has this jurisdiction actually done with similar applications."

5. Fiber and connectivity. Which carriers and long-haul routes are present or nearby, and whether the site can plausibly obtain the diverse routing a data center requires. A site with power but no realistic path to diverse fiber is a stalled project with extra steps.

6. Water and cooling. The water supply situation matched against the cooling strategy from step zero: municipal or well supply, any known allocation constraints, and whether an air-cooled or liquid-cooled design changes the answer. Water is screened as a real constraint, not a footnote, because in several regions it is the constraint that arrives last and kills fastest.

7. Environmental and hazard exposure. Publicly mapped risks: flood zones, wildfire severity areas, seismic considerations, wetlands and protected habitat indicators, and any record of prior industrial use that makes a Phase I environmental assessment likely to find something. The screen uses public mapping and records; it does not sample soil or groundwater, and it says so.

8. Development potential and the diligence plan. Pulling the first seven areas together: what the site can realistically support (in MW, by phase, with stated assumptions), what it cannot, and the specific next steps required to move from screening conclusions to diligence-grade certainty.

The capacity ladder: four states, never mixed

Power claims are where screening most often goes wrong, because a single megawatt number can mean four very different things. We classify every capacity figure into exactly one state:

• Planned — announced, contemplated or shown in a plan. No utility filing or commitment exists.
• Applied — a formal interconnection or large-load service application has been filed and is under study. Nothing has been granted.
• Committed — the utility has completed its study process and committed capacity in writing, normally with an agreement, a delivery schedule and customer financial obligations.
• Operational — built, energised and in service.

The distinctions are not pedantry; the public data supports them. Berkeley Lab's Queued Up: 2024 Edition found that of roughly 2.6 TW of generation and storage capacity in U.S. interconnection queues at the end of 2023, only about 12% (311 GW) had reached executed interconnection agreements — the point at which a request becomes a commitment — and historically only about 14% of queued capacity from 2000–2018 had actually been built. That dataset covers generation and storage rather than data center loads, but the structural lesson transfers directly: an application is not a commitment, and a commitment is not an operating asset. A Site Check report that writes "100 MW" without stating which rung of the ladder it sits on, and what document establishes that, has not answered the power question — it has repeated a marketing claim.

Evidence discipline: three kinds of statement

Every statement in a Site Check report carries one of three labels:

• Confirmed fact — supported by a named source, with its date. Utility territory maps, recorded parcel data, adopted zoning code, a utility's written response.
• Reasonable inference — our professional reading of confirmed facts, labelled as inference. For example: a substation of a given configuration adjacent to a parcel supports an inference about feasible capacity ranges; it does not confirm any specific capacity.
• To be verified — material questions the screen could not answer, stated openly with the step that would answer them.

Sources are recorded with the date they were captured, because utility positions, zoning and queue status all change. A report that does not show its sources and their dates cannot be audited, and its conclusions should not be relied on.

Kill criteria

A screen earns its cost when it stops bad projects early. In our method, any of the following ends the evaluation or escalates it to expert review before further spending: no credible path to the required capacity in any of the four states within the project's timeline; zoning that prohibits the use with no documented amendment path; a hazard exposure (flood, wildfire, seismic) that insurance or design cannot reasonably absorb; water requirements the local supply demonstrably cannot meet; or title defects public records already reveal. A "fail" from a Site Check is a useful, money-saving result — precisely the outcome the screen exists to produce cheaply.

What a Site Check is not

Three limits are stated in every report we produce. First, a Site Check is a screen, not diligence: it does not include a utility load study, a title commitment, a Phase I environmental site assessment, geotechnical work, or a formal legal opinion, and it must never be represented as any of those. Second, it is not a professional opinion of record: where a licensed engineer's, surveyor's or attorney's judgement is required, the screen's role is to identify that requirement, not to satisfy it. Third, it expires: utility capacity positions and regulatory status change, so screening conclusions carry a stated review date.

Used properly, the sequence is simple: screen first, in writing, with sources — then spend the diligence budget only on sites that pass. With CBRE reporting North American vacancy at a historic low of 1.4% at the end of 2025 and describing a "power-first" market, the cost of discovering a power problem late is measured in lost position, not just lost fees.

Sources

1. Lawrence Berkeley National Laboratory, Queued Up: 2024 Edition — Characteristics of Power Plants Seeking Transmission Interconnection, April 2024. ~2.6 TW active in queues at end of 2023; ~12% (311 GW) with executed interconnection agreements; ~14% of 2000–2018 queued capacity built. Reported by Utility Dive, April 11, 2024. https://www.utilitydive.com/news/grid-interconnection-queue-berkeley-lab-lbnl/712926/ — Evidence grade: A/B.
2. CBRE, North America Data Center Trends (H2 2025), press release "Fast-Growing North American Data Center Market Set Records in 2025," 2026. Vacancy of 1.4% (historic low) at end of 2025; "power-first approach that prioritizes sites with the fastest path to power." https://www.cbre.com/press-releases/fast-growing-north-american-data-center-market-set-records-in-2025 — Evidence grade: B.
3. AIDCKOL, Site Check methodology (internal framework, first published in this article, 2026-10-23). The eight screening areas, four-state capacity classification and three-tier evidence labelling are AIDCKOL's own method, presented as such. — Evidence grade: not applicable (method statement, not a factual claim).