Build a state shortlist for your next datacenter project. Compare power, policy and operating conditions for hyperscale and AI training facilities, then identify what to verify locally.
Choose a size band and a priority preset, then adjust the factors that matter to your project.
Open state profiles, save candidates, and compare their tradeoffs in the table.
Use the shortlist to scope utility, permitting and water diligence. State fit does not establish site readiness.
States farther right score better on power, water, hazards and cooling. States higher up score better on policy and permitting. Dashed lines mark the median scores.
The axes use reference weights; dot colors show fit in the selected view. Priority sliders change colors, while power cost assumptions also affect horizontal positions. Select a dot for its profile. Scroll across the chart on small screens.
Sort by any column and select a row for details. Judgment ratings run from 1 to 5, and fit and reference scores run from 0 to 100. Delivery concerns are a separate modeled signal. Higher is more favorable, except for electricity prices.
Pushback describes public opposition; rules in force summarize statewide regulation. Neither affects fit or rank. The Reception tab provides details.
Saved for this visit. Ranks stay relative to all 50 states. Export includes the displayed states, active weights and model assumptions.
EIA’s September 9, 2026 outlook forecasts growing US electricity sales, driven by datacenter development and manufacturing. These national forecasts provide context for utility diligence; they do not measure spare capacity in a state.
The commercial sector includes more than datacenters. TWh means billion kilowatt-hours. All figures are forecasts, not observed demand or datacenter-only consumption, and contribute no points to state fit.
Use in diligence: ask the serving utility how its load forecast affects your proposed connection date and upgrade requirements.
Source: EIA Short-Term Energy Outlook, September 9, 2026 · Checked September 13, 2026. The source page updates monthly; figures here retain the September release.
This exploratory model reports weighted suitability, delivery concerns and rating sensitivity. Public response and regulation are reported separately and do not affect scores. The model does not predict project approval.
Balanced starts with equal importance for the eight grid-siting factors: power headroom, power cost, incentives and policy, community and permitting, water, natural hazards, cooling climate, and connectivity. Each contributes 12.5% initially. The ninth factor, winter solar resource, starts at zero because a grid-connected project buys power rather than generating it. Presets and sliders change those shares.
Fit = sum of each factor's score × its share. Seven factors retain provisional analyst ratings on a 1–5 scale, mapped to 0, 25, 50, 75 and 100. These equal steps are a modeling assumption, not measured differences. Momentum is shown only as context: existing development can signal an ecosystem, but can also consume power capacity.
Power price and winter solar resource are the two measured factors. By default, the cost score declines linearly from 100 at 5¢/kWh to 0 at 35¢/kWh, with scores capped at those endpoints. These limits distinguish all prices in the current dataset. Change them under Power cost assumptions to test your preferences. They are not forecasts or tariffs.
Winter solar resource is the December mean all-sky irradiance from NASA POWER's 2001–2020 climatology, read at each state's 2020 Census center of population, in kWh per square metre per day. December sizes an off-grid array and battery, so it stands in for the year. The score rises linearly from 0 at 1.0 to 100 at 4.0 kWh/m²/day, which spans the December values from Washington and Vermont near 1 to Hawaii near 4. One grid cell of about half a degree per state is a coarse proxy: a state's sunniest and cloudiest sites can differ by a third or more, and the value says nothing about land, interconnection or permitting for a solar array.
Four size bands describe the project: edge or micro (under 5 MW), enterprise or colocation (5 to 50 MW), hyperscale (50 to 300 MW) and gigawatt campus (300 MW and up). Choosing a band loads a starting weight profile, which the presets and sliders then adjust. Edge starts from connectivity 40%, hazard 15%, cost 15%, climate 10%, permitting 10%, headroom 5%, policy 5%. Enterprise matches Connectivity first. Hyperscale starts from headroom 25%, cost 20%, policy 15%, permitting 15%, water 10%, hazard 5%, climate 5%, connectivity 5%. Gigawatt starts from headroom 35%, cost 25%, permitting 20%, policy 10%, water 5%, hazard 5%, with climate and connectivity at zero.
Size also sets the headroom and permitting ratings a project needs before its delivery concerns read as lower: edge needs 2 and 2, enterprise 3 and 2, hyperscale 3 and 3, gigawatt 4 and 3. These thresholds are analyst judgments about scale, not measured capacity, and a state's ratings do not change with size. The page opens on hyperscale with Balanced weights.
A fifth band, off-grid solar and battery, describes a self-powered unit of any size. It starts from winter solar resource 35%, community and permitting 20%, natural hazard 15%, cooling climate 15%, connectivity 10% and incentives 5%, with power cost, headroom and water at zero, since the unit buys no grid electricity, joins no interconnection queue and is usually air cooled. Only permitting can raise its delivery concern, at 3 of 5. Three workload buttons move the connectivity weight alone: batch inference 10, real-time serving 25, distributed training 40, on the same relative scale as the sliders. Battery autonomy, land cost, local maintenance capacity and physical security are site questions with no state-level public source and are not modeled.
Power headroom and community/permitting are each compared with the size threshold. A rating at or above the threshold means lower concerns, one level below means elevated concerns, and two or more below means major concerns. The worse of the two factors sets the state's level. For hyperscale that is the original rule: a rating of 1 means major, 2 elevated, 3–5 lower. Lower concerns do not establish available power or permit eligibility.
Weighted fit permits trade-offs and is the default. Delivery first ranks lower-concern states ahead of elevated- and major-concern states, then compares fit within each group. Cheap power therefore cannot offset a worse delivery group in that mode. The grouping still relies on provisional ratings, not verified connection timelines.
Each profile's Scores tab shows the best and worst rank obtained by changing one of that state's judgment ratings by one level, within 1–5. Other ratings and states stay fixed. This is a limited sensitivity test, not a confidence interval or probability. It does not capture correlated errors or simultaneous changes across states.
Exact score ties within the selected ranking rule share a rank. Setting every weight to zero removes fit scores and ranks while keeping facts and sources available.
The page address records the size, priorities, ranking rule, price anchors, figure and shortlist, each omitted when it is at its default. Copy scenario link copies that address. Download share image draws a 1200 by 675 pixel card of the five leading states and the priority shares for the current scenario. Both describe this model's output, not a site decision.
The Reception tab reports public response and statewide regulation. The table summarizes opposition and the number of regulatory categories with requirements in force. Neither affects scores or rankings.
Pushback is Low where no organized opposition or local restriction is on record, Moderate where local restrictions or contested proceedings exist in some jurisdictions, and High where there are multiple local bans, a statewide restriction that cleared a chamber, or a referendum. Local action counts here. The four rule axes stay strictly statewide and cover water, power and ratepayer cost, siting and zoning, and tax incentives.
The four axes classify statewide rules in force: entry facilitation, no specific rule, disclosure or reporting, binding conditions, and pause or prohibition. Pending bills are described in the notes and do not affect these classifications. Tax incentives use corresponding eligibility categories. Classifications are provisional and have not been individually verified against original records.
Public opposition and regulatory requirements are distinct: substantial opposition may coexist with permissive rules, and detailed regulation may coexist with support for development.
Political outlook weights policy and permitting equally. Physical fundamentals weights power headroom 30%, price 25%, water 20%, hazards 15% and climate 10%. These reference weights stay fixed. Editing a priority returns to Combined; changing the cost scale also affects the physical reference score.
All 50 historical electricity prices were compared with EIA's 2024 industrial data. Solar irradiance comes from NASA POWER at one point per state, with eight states spot-checked against NREL's solar resource service. The seven judgment factors have not been individually validated. A state citation can support a policy event without proving its numerical rating.
State profiles distinguish government records, grid publications, reporting, legal analysis, market research, company statements and advocacy statements. Company plans are self-reported; an advocacy organization's account of a lawsuit is not a court finding. A bill or agenda item is not automatically an enacted restriction.
Commercial tracker totals and references without identifiable documents have been removed from the displayed source list. Some unsupported counts and subsidy estimates have also been removed from state summaries. Remaining summaries and ratings are provisional and still need a claim-by-claim review.
Model version 3 · Evidence review September 7, 2026 · Solar factor and off-grid band added September 23, 2026 · County, utility territory and project-specific conditions require separate investigation.
The dataset covers 50 states: electricity prices, December and annual solar irradiance, seven scored analyst ratings, momentum, policy posture, reception and regulation, and citations. CSV provides one row per state; JSON preserves the source structure.
Openly licensed under CC BY 4.0, and also in the public repository. Suggested citation: Sonnet Xu, US datacenter comparison, 2026. No registration required.