Data Center Site Selection Criteria: The 11 Factors That Decide Buildability in 2026

In 2020, the site selection question was "where is land cheap and fiber close?" In 2026, it's "where can I actually get 300 MW before 2030?"

That shift changed the entire criteria list. Fiber, land cost, and tax incentives still matter — but they no longer decide outcomes. Power delivery does. In ERCOT alone, roughly 226 GW of large loads have asked to connect, while only about 5.3 GW are actually drawing power.^[1] Utilities have responded by tightening their queues: AEP cut its large-load forecast from roughly 30 GW to 13 GW after requiring real financial commitments,^[2] and nearly a quarter of Georgia Power's large-load pipeline exited when deposits came due.^[3]

The practical takeaway: most announced capacity never becomes delivered capacity. Site selection in 2026 is the discipline of separating the sites that can be energized from the sites that merely look available.

Here are the 11 criteria that determine buildability — ranked by how often each one actually kills a project.

Why the criteria list changed

Three forces rewrote the checklist. First, hyperscale and AI demand pushed single-site requests from 30–60 MW to 300 MW and beyond, which means almost every serious site now requires transmission-level service and formal utility study. Second, interconnection queues became the binding constraint — time-to- power in constrained markets now runs 2 to 7 years, and queue position is often worth more than the land under it. Third, community opposition matured from scattered complaints into organized moratoria, restrictive ordinances, and ballot outcomes that can end a project after millions in diligence spend.

The result is that the traditional criteria list — connectivity, climate, cost — is now table stakes. The differentiating criteria are the ones most published checklists treat as a single bullet: power.

The 11 criteria, ranked by 2026 impact

1. Power availability and interconnection timeline

The first question is not "is there a substation nearby?" It's "what can that substation actually serve, and when?" Proximity to infrastructure is not the same as access to capacity. Evaluating this criterion properly means looking at indicative available capacity at the serving bus, the site's realistic position relative to the utility's study process, network- upgrade exposure, and the milestones between today and an executable agreement.

This is where most sites fail — and where the difference between sites is largest. Upgrade-cost exposure can swing an order of magnitude between two parcels that look identical on a map. A site with a credible 24-month path to 100 MW is a fundamentally different asset than a site with a speculative 6-year path to 300 MW, even if the second one's marketing brochure looks better.

What to demand as evidence: utility correspondence (not broker representations), study status with dates, and a stated basis for any capacity claim. "Grid capacity confirmed" without a completed system impact study is a representation, not a fact.

Study timing itself has changed. Grid operators are moving from serial, first-come-first-served study processes to batched or clustered windows — ERCOT's Batch Zero process for large loads took effect in July 2026, and cluster-style approaches govern study cycles in other regions. The practical consequence: interconnection timing is now cohort-based. Miss a submission window and the site doesn't slip a few weeks — it waits for the next cycle.

There is also now a second path to power. Behind-the-meter and bring-your-own-generation strategies — on-site generation, co-location with existing plants, bridge power ahead of grid service — can compress time-to-power meaningfully. But they trade a queue problem for a commercial-complexity problem: offtake structure, backup and standby service, interconnection for export, and exit rights all have to be negotiated across parties with different incentives. BYOG is a legitimate answer to this criterion for the right site; it is not a shortcut around diligence.

2. Utility willingness and large-load tariffs

The tariff a large load signs under now shapes project economics as much as the energy price itself. Utilities across the country are introducing large-load tariffs with minimum- demand commitments, contracted-capacity clauses, collateral requirements, and take-or-pay structures — terms designed to protect other ratepayers from stranded investment when announced projects don't materialize. AEP's shift to requiring minimum payment commitments is exactly why its pipeline dropped from 30 GW of requests to 13 GW of credible demand.^[2]

Two sites in adjacent service territories can face materially different obligations. A utility that has built a clear large- load process and wants the load is a different counterparty than one facing a moratorium debate at its commission. This criterion barely appears on most published checklists — and it is frequently the difference between a financeable deal and a stalled one.

Two shifts are reshaping this criterion in real time. First, flexibility has become currency: loads that can accept curtailment during system peaks are increasingly offered faster interconnection paths — ERCOT's large-load interconnection standards and SPP's expedited high-impact-load processes both reward controllability. A load that can flex is a different negotiation than a load that can't. Second, cost allocation is moving. Texas regulators have approved a rulemaking replacing ERCOT's 4CP transmission cost-allocation mechanism with a 12CP methodology^[4] — which erodes the peak-avoidance economics some large loads have relied on and makes transmission charges a larger, less avoidable line item in the pro forma. Underwrite the tariff you'll actually sign, not the one the last deal signed.

3. Community sentiment and permitting risk

Community opposition is now a first-order site risk, not a public-relations footnote. Counties and municipalities have enacted data center moratoria, restrictive zoning overlays, and noise, water, and setback ordinances — and projects have died at the rezoning stage after significant diligence investment. Opposition tends to concentrate where hosting burden is already high and local benefit is perceived as low.

Assess it like any other technical criterion: what has the jurisdiction already enacted, what is pending, what happened to the last comparable project, and how does this county's hosting load compare to its local demand?

4. Fiber and network latency

Connectivity has shifted from a top-three criterion to a qualifier. Long-haul routes and middle-mile builds have expanded enough that most power-viable sites can be served — at a price. Verify route diversity (two truly independent paths, not two providers on the same conduit), realistic construction cost for laterals, and latency requirements specific to the workload. Training-oriented campuses tolerate latency that inference and colocation workloads cannot.

5. Climate and cooling economics

Cooling drives a meaningful share of operating cost, and climate drives cooling. Evaluate wet-bulb temperature profiles, annual free-cooling hours, and — critically — whether local water constraints will force the design toward air-cooled or closed-loop systems that change the power and capex math. A site with cheap power but a hostile cooling climate can lose its economic edge quickly at hyperscale.

6. Water access and rights

Water is now both an engineering input and a political one. Even designs with modest consumption face scrutiny in stressed basins, and water availability is among the most common drivers of community opposition. Confirm the legal right, the physical delivery path, and the political durability of both. In water-stressed regions, plan for the design that assumes water is not available.

7. Land: cost, topography, contiguity

Land cost matters least of the criteria on this list — it's typically a low-single-digit share of total project cost — but land quality matters a great deal. Contiguity for phased expansion, developable acreage after setbacks and easements, geotechnical conditions, floodplain exposure, and clean title with control instruments that survive a multi-year development timeline. A cheap parcel with a pipeline easement through the buildable envelope is not cheap.

8. Natural-disaster and long-horizon climate exposure

Underwrite the site for a 20–30 year operating life, not current conditions: seismic zone, tornado and hurricane exposure, wildfire risk, and hundred-year flood mapping that is increasingly out of date. Insurance markets are already repricing these risks; your pro forma should too.

9. Tax incentives and abatements

Most competitive states now offer sales-tax exemptions on equipment, and many offer property-tax abatement structures for qualifying investment. Incentives are real money, but they are the last criterion to optimize, not the first — an incentive package cannot fix a site that can't be powered. Treat published incentive thresholds (minimum investment, job counts, wage floors) as diligence items, and note that several states are actively re-examining data center incentives as ratepayer debates intensify.

10. Workforce availability

Construction workforce is the binding constraint more often than operating workforce. A hyperscale build can require thousands of tradespeople at peak; in markets with multiple simultaneous builds, labor scarcity is extending schedules and inflating cost. Operating staff needs are modest by comparison but favor markets with existing mission-critical, industrial, or military-adjacent talent.

11. Zoning and entitlement path

Last on the list because it's the most fixable — but only with time, and time is the scarcest resource. By-right industrial zoning is worth a premium over any site requiring rezoning, because rezoning is where community sentiment (criterion 3) gets a formal vote. Map the full entitlement sequence — zoning, site plan, utility easements, road and traffic approvals — against the power timeline from criterion 1. The site is only as fast as its slowest approval.

How the criteria vary by market

The weighting above is national. The binding constraint shifts by region:

PJM

The deepest data center market and the most congested queue conversation in the country. Capacity market outcomes and large-load cost-allocation debates are actively reshaping the economics; established data center corridors carry both the best infrastructure and the most organized opposition.

ERCOT

The most flexible market for large loads — and the one that has moved fastest to impose discipline on speculative requests, including new interconnection standards for large loads. The gap between requested and energized capacity is the defining fact of this market: interconnection process knowledge is worth more here than anywhere.^[1]

MISO

A broad footprint with genuine capacity headroom in parts of the region, structured study cycles, and growing large-load activity. Utility-by-utility variation in large-load readiness is wide — the serving utility matters more than the RTO label.

SPP

Lower land and power costs with real available capacity in portions of the footprint, balanced against longer transmission distances and thinner fiber in rural areas. An underrated market for loads that can trade latency for cost and speed.

Southeast (non-RTO)

Vertically integrated utilities mean one counterparty controls generation, transmission, and the timeline — which can be the fastest path in the country or the slowest, depending entirely on that utility's posture and headroom. Georgia's pipeline attrition when deposits came due shows both the scale of speculative demand and the value of showing up as a credible counterparty.^[3]

The pattern across all 11 criteria

Every criterion on this list reduces to the same question: what's the difference between what's represented and what's proven? A capacity claim without a study, a water right without a delivery path, an incentive without a qualifying threshold review, a zoning assumption without an entitlement map — each is a gap between potential and proof.

That's the discipline GridMatch was built around. Our GRIP℠ methodology scores sites across 12 diligence categories and discounts every input by the quality of its evidence — so a validated utility confirmation counts for more than a broker representation, and the composite score reflects what's actually known, not what's hoped. Sites that screen well move into full development advisory through GROUND℠ and utility engagement through GUIDE℠.

Learn how GRIP℠ scores a site →

Frequently asked questions

Related resources