Power, not land, now decides where America's data centers get built. Utility-scale electricity demand from U.S. facilities is on pace to jump from 31 gigawatts in 2025 to 41 GW this year and 66 GW by 2027, according to Goldman Sachs Research — a curve steep enough that interconnection queues, not zoning maps, have become the real gating factor for developers.
On this page
- Power availability & grid capacity
- Fiber connectivity & latency
- Economic incentives & tax exposure
- Land, zoning & permitting speed
- Climate, cooling & free-air economics
- Natural hazard & resilience risk
- Water resources & sustainability exposure
- Labor market & logistics access
- Total cost of ownership modeling
- Five shifts reshaping site selection in 2026
- Comparison table: leading U.S. markets
- Advisory for stakeholders
- FAQ
Selecting a U.S. data center site has stopped being a real estate decision that happens to involve utilities, and become an energy-procurement decision that happens to involve real estate. The order of operations has flipped: developers now lock in a power agreement first and search for land second, a reversal that is reshaping which metros win and which get passed over entirely. Below is a sector-by-sector breakdown of the criteria driving 2026 site decisions, grounded in the latest utility, grid-operator, and market data.
Power Availability and Grid Capacity
Power remains the single deciding factor in U.S. site selection, and the gap between forecast demand and deliverable capacity is widening rather than closing. Goldman Sachs Research now projects U.S. data center power demand climbing from 31 GW in 2025 to 41 GW in 2026 and 66 GW in 2027, implying installed capacity of roughly 95 GW by the end of 2027 — more than double the 2025 level. Separately, S&P Global's 451 Research puts total utility power supplied to hyperscale, leased, and crypto-mining facilities at about 64.4 GW in 2025, up 25% year over year, and forecasts that figure will nearly triple again by 2030 to roughly 183 GW.
The bottleneck sits at the grid edge, not inside the fence line. Interconnection queues in constrained regions such as the Mid-Atlantic, Mid-Continent, and Northwest now stretch out several years, and transformer lead times exceeding 128 weeks mean that capacity announced today often can't be energized until the back half of the decade. Analysts also caution that reported demand is somewhat inflated: developers routinely file interconnection requests with multiple utilities for the same project, so utility load forecasts submitted to grid operators may overstate near-term additions by tens of gigawatts.
Virginia remains the largest single market for data center grid power, with demand forecast to rise 27.7% in 2026 to 16.6 GW, according to 451 Research. But Texas is closing the gap fastest, projected to grow 35.9% to 13.5 GW this year, while Ohio and Iowa are each expected to post growth rates above 50%, reaching 5.3 GW and 3.5 GW respectively. That divergence — a mature hub decelerating while frontier markets accelerate — is the clearest signal of where the next wave of site selection decisions is heading.
Developers have adapted by securing power purchase agreements and interconnection rights before touching land acquisition — the opposite sequence from a traditional build-to-suit real estate deal, and now the standard playbook for anything above 50 MW.
Fiber Connectivity and Network Infrastructure
Latency-sensitive workloads still anchor site selection to dense fiber corridors. Proximity to internet exchange points and carrier-neutral routes determines both network performance and the pool of tenants willing to lease space. Ashburn, Virginia remains the reference case: its concentration of long-haul fiber and internet exchange points gives it sub-millisecond connections to much of the East Coast, which is why "Data Center Alley" still commands premium rents even as power constraints push new construction elsewhere.
Economic Incentives and Tax Exposure
More than 30 states now offer data center–specific incentives — typically sales-tax exemptions on servers and cooling equipment, property-tax abatements, and direct grants — and in competitive site searches these have shifted from a bonus to a baseline requirement. Texas and Virginia built early leads by pairing aggressive incentive packages with strong grid access, but the incentive landscape has become more contested, not less, heading into 2026.
That contest now cuts both ways. In 2025, more than 200 bills addressing data centers were introduced across all 50 states and over 40 were signed into law, according to the Information Technology and Innovation Foundation — many aimed at cost-shifting protections for ratepayers rather than new subsidies. A bipartisan coalition of governors, including Virginia's Abigail Spanberger, has pressed PJM Interconnection to shield residential customers from data-center-driven grid costs, a sign that incentive generosity is increasingly being weighed against political tolerance for higher electricity bills.
Land Availability, Zoning, and Permitting Speed
Industrial-zoned parcels with existing utility infrastructure remain the preferred footprint, but permitting velocity has become a competitive differentiator in its own right. States such as Arizona and Nevada continue to streamline approvals relative to more heavily regulated markets, and speed-to-market advantages compound: a project that reaches energization twelve months earlier captures a full extra year of lease revenue before the next supply wave arrives.
Local pushback is the countervailing force. Moratoriums or outright bans on new data center construction are under active discussion in more than 20 states, and industry trackers estimate over $130 billion in projects were delayed or abandoned in the first quarter of 2026 alone amid community and grid-cost opposition — a reminder that entitlement risk in 2026 is as much political as procedural.
Climate, Cooling, and Free-Air Economics
Cooler, drier climates continue to command a real operating-cost premium because they extend the number of hours per year a facility can rely on free-air cooling instead of mechanical chillers. The Pacific Northwest remains the benchmark region for this reason, pairing low ambient temperatures with access to hydropower. But the calculus is shifting as liquid cooling becomes standard for AI training clusters: liquid-cooled racks are less sensitive to outdoor climate than legacy air-cooled halls, which is gradually loosening climate's grip on site selection in favor of power and water access.
Natural Hazard and Resilience Risk
Uptime requirements approaching 99.999% leave little tolerance for site-level hazard exposure. Flood zones are generally avoided outright rather than mitigated, since even a low-probability event can trigger tenant default clauses and insurance non-renewal. Wildfire, hurricane, and seismic exposure now factor directly into capitalization rates: two otherwise identical sites can carry a meaningfully different cost of capital once an insurer prices in geographic risk.
Water Resources and Sustainability Exposure
Water has moved from an operational footnote to a headline permitting risk. U.S. data centers directly consumed an estimated 17.4 billion gallons of water in 2023, and researchers at Lawrence Berkeley National Laboratory project direct consumption could reach 38 to 73 billion gallons by 2028 as AI training and inference workloads push more sites toward evaporative and liquid cooling. Two-thirds of new U.S. hyperscale campuses built since 2022 sit in counties already classified as high or extreme water-stress, according to Bloomberg reporting cited in recent policy research — a mismatch that is now driving permit denials rather than just community complaints.
Not every operator is moving at the same pace or with the same footprint. Amazon disclosed for the first time in June 2026 that its global fleet consumed 2.5 billion gallons of water in 2025 at a water-usage-effectiveness ratio of 0.12 liters per kilowatt-hour, driven largely by free-air cooling. Google's 2026 environmental disclosure showed 10.9 billion gallons consumed in 2025, up 34% year over year and more than double its 2021 level, reflecting the water intensity of its AI-optimized facilities. The spread between those two figures is itself becoming a site-selection variable, since jurisdictions are starting to write water-usage-effectiveness thresholds directly into permitting requirements.
Labor Market and Logistics Access
Data centers are not labor-intensive once operational, but construction-phase staffing and specialized maintenance technicians are increasingly scarce in the same secondary markets now attracting hyperscale capacity. Metros such as Dallas and Atlanta continue to benefit from deep skilled-trade labor pools and strong highway and air-freight access for the specialized equipment — generators, transformers, cooling units — that a large campus requires.
Total Cost of Ownership: The Metric That Actually Decides
Land price is the smallest line item in most site comparisons and often the least decisive. Total cost of ownership blends energy pricing, tax exposure, labor, water, and — increasingly — the cost of capital penalty attached to hazard or reputational risk. A parcel with a higher acquisition cost but cheaper, more reliable power and a faster permitting path routinely outperforms a "cheap land" alternative once modeled over a 15-to-20-year hold period.
Five Shifts Reshaping Site Selection in 2026
1. The center of gravity is moving to secondary markets
Grid constraints in legacy hubs are pushing new capacity toward Ohio, Indiana, and other markets where interconnection queues are shorter, even if fiber density is lower.
2. Co-location with dedicated generation
On-site gas turbines, small modular reactor commitments, and direct renewable off-take agreements are increasingly bundled into the site decision itself rather than sourced afterward.
3. AI-assisted site screening
Developers are using machine-learning tools to score thousands of parcels against power, fiber, and hazard data simultaneously, compressing a process that used to take quarters into weeks.
4. Water-usage-effectiveness as a permitting gate
Following Minnesota's 2025 water-permitting requirement for large data centers, expect more states to require a disclosed WUE figure before issuing a construction permit.
5. Political risk is now underwritten explicitly
With over 40 new state laws enacted in 2025 and moratorium debates active in 20-plus states, legal and government-affairs due diligence has become as central to site selection as engineering due diligence.
Comparison Table: Leading U.S. Data Center Markets, 2026
| Market | 2026 Power Demand | YoY Growth | Primary Advantage | Primary Constraint |
|---|---|---|---|---|
| Virginia (Northern VA) | 16.6 GW | +27.7% | Fiber density, IXP proximity | Grid congestion, ratepayer politics |
| Texas | 13.5 GW | +35.9% | ERCOT flexibility, incentives | Water stress, summer peak risk |
| Ohio | 5.3 GW | >50% | Shorter interconnection queues | Lower fiber density |
| Georgia | 4.1 GW | +34.7% | Labor pool, logistics | Rising interconnection requests |
| Iowa | 3.5 GW | >50% | Cool climate, wind power access | Smaller labor market |
| Indiana | 1.6 GW | >100% | Land availability, incentives | Early-stage fiber build-out |
Sources: 451 Research / S&P Global (2026 forecasts); Goldman Sachs Research; ITIF policy tracking.
Advisory for Stakeholders
The 2026 site-selection environment rewards stakeholders who sequence their diligence correctly and penalizes those who treat power, water, and political risk as afterthoughts. The following is directed guidance by stakeholder type.
For developers and institutional investors
Underwrite interconnection timing before land price. A site with a 2028 energization date is a different asset class from one that can go live in 2026, even at identical acreage cost. Build water-usage-effectiveness disclosure into your own reporting now, ahead of state mandates — it will shorten permitting timelines rather than lengthen them once regulators start requiring it anyway. Diversify across at least two power markets; concentration in a single constrained grid (Mid-Atlantic, in particular) now carries measurable reliability risk per Goldman Sachs Research.
For utilities and grid operators
Treat duplicate interconnection filings as a data-quality problem, not just a queue-management one; analysts estimate double-counted requests may inflate near-term load forecasts by roughly 25 GW nationally. Ratepayer-protection tariff structures — separating data-center-specific infrastructure costs from general rate base — are moving from advocacy talking point to enacted policy in multiple states and should be modeled into 2026-2027 rate cases now rather than reactively.
For state and local governments
Incentive packages without accompanying water and grid-cost safeguards are increasingly a political liability, evidenced by the 40-plus new state laws passed in 2025 and moratorium debates active in over 20 states. Jurisdictions that pair fast permitting with clear water-consumption and ratepayer-protection standards — rather than choosing one or the other — are best positioned to keep capturing investment without triggering the backlash now stalling projects elsewhere.
For hyperscale tenants and enterprise buyers
Require landlords to disclose fiber-path redundancy and grid interconnection status in writing before signing, not just power capacity figures. Given that two-thirds of new hyperscale campuses sit in water-stressed counties, tenants with public sustainability commitments should independently verify a site's water-usage-effectiveness rather than relying solely on developer marketing claims.
For communities and ESG-focused stakeholders
Ask for basin-specific water data, not fleet-wide averages — a withdrawal from a stressed aquifer carries different risk than the same volume from a water-abundant watershed, and hyperscalers now disclose this at the facility level. Track ratepayer-impact studies alongside job-creation projections when evaluating a proposed project; both are now publicly available in most state utility filings.
Conclusion
Data center site selection in 2026 has fully inverted from the model that governed the last two decades: power availability, interconnection timing, and water-permitting risk now outrank land price and even fiber density in determining where hyperscale capacity actually lands. The developers, utilities, and governments moving fastest are the ones that have already restructured their diligence around that inversion — securing energy first, modeling total cost of ownership over a full hold period, and pricing in political and water risk explicitly rather than treating them as afterthoughts.
Related reading: Commercial Asset Management in Residential Developments: Revenue Share Models
Frequently Asked Questions
What is the single most important factor in 2026 data center site selection?
Power availability and interconnection timing, by a wide margin — Goldman Sachs Research projects U.S. demand reaching 41 GW in 2026 and 66 GW in 2027, and queue delays now determine project viability more than any other single variable.
Which U.S. markets are growing fastest right now?
Texas, Ohio, Iowa, Indiana, and Georgia are posting the steepest year-over-year growth in 2026, per 451 Research, even though Virginia still holds the largest absolute demand.
How long does site selection and energization typically take?
Site evaluation itself typically runs six to eighteen months, but power interconnection can add another two to five years in constrained grid regions.
Is water becoming a bigger constraint than power?
Not yet nationally, but it is now a binding constraint in specific basins — Texas alone is projected to see data center water use rise from roughly 49 billion gallons in 2025 to as much as 399 billion gallons by 2030.
Core Insights Review's editorial team covers commercial real estate, PropTech, smart infrastructure, sustainable construction, industrial real estate, and the technologies shaping the built environment.
Check for more information: Core Insights Review
