Data centers · published 2026-07-05 · 6 min read
Site Selection Is Now Grid Selection: Screening Land for Data Centers
By Shashwat Kapoor
Photo: Christopher Down, CC BY 4.0, via Wikimedia Commons.
For a data center, power is now the constraint that decides everything, and the winners are the teams that can screen sites faster than the queue and the market move against them. Doing that well means pulling power, fiber, water, and climate data for a single coordinate in seconds, when today it can take weeks across a dozen disconnected systems.
North American data center vacancy has fallen to roughly 1.9%. Prime land in Northern Virginia now trades above $2 million per acre, and in power-constrained markets, time-to-power runs a baseline of 24 to 36 months before a single server is energized. A single mis-sited campus can strand hundreds of millions in capital and push a commercial-operations date years past the window it was underwritten against.
For most of the last decade, fiber density and interconnection gravity decided where facilities landed. That logic has inverted. Power deliverability now sets the boundary of what is possible, and the industry has settled on a phrase for it: site selection is now grid selection.
The criteria themselves are no secret. Every experienced developer can recite them. The advantage now comes from clearing those criteria as a sequential filter, on data you can trust and defend, fast enough to secure power before the queue fills and the window closes. What follows is that filter, and the data discipline that has to sit underneath it.
Power and Energy Strategy: The Binding Constraint
Speed to power. That's the name of the game. It decides whether a project gets built inside its schedule, and it sits first because a site that fails on power cannot be rescued by strong fiber, cheap land, or a generous incentive. Everything downstream is contingent on it.
Photo: Matthew T. Rader, CC BY-SA 4.0, via Wikimedia Commons.
Time-to-power outranks nameplate megawatts
The relevant question is timing. Ask when firm power can actually be energized at the voltage class you need, and treat capacity on a map as a weak proxy for that answer. Both JLL and CBRE now rank speed-to-power as the primary greenfield criterion, a marked change from the fiber-first logic of the last cycle.
That ranking is redrawing the map. Tier-1 hubs such as Northern Virginia and Santa Clara are congested by grid limits, multiyear permitting, and land prices past $2 million per acre, so hyperscalers are pivoting to Tier-2 markets like Columbus, Des Moines, and San Antonio, where power can arrive 12 to 24 months faster and land runs up to 70% lower.
Interconnection queue: proximity is not deliverability
Being near a substation or a transmission line is a shallow signal. What matters is whether the grid operator can study, approve, upgrade, and serve your load inside your schedule. U.S. interconnection queues now hold on the order of 2,300 GW of capacity, and average wait times have climbed from under 2 years in 2008 to roughly 5 years today, past 9 years in California.
Before you sign an LOI, pressure-test three things: available substation headroom at your voltage class, a realistic energization date, and the bridge plan if that date slips. The picture is also dynamic, since a single nearby battery, solar, or hyperscale load request can trigger restudies and consume the upgrade capacity you were counting on.
Securing firm power: PPAs, behind-the-meter, and co-located generation
When the public queue cannot serve the schedule, on-site or adjacent generation becomes the fastest route to firm power. Natural gas paired with battery storage is the common bridge in constrained markets, and proximity to an existing gas pipeline often decides whether that bridge is even viable. Behind-the-meter arrangements carry a second benefit, since they insulate a project from congestion pricing and the volatility of interconnection-upgrade costs.
A power purchase agreement is an infrastructure-access tool as much as a sustainability instrument, and the right structure can fund new generation or even bypass the public queue entirely. A virtual PPA suits creditworthy buyers matching renewable attributes across markets, a physical or sleeved PPA ties a specific project to the facility, and co-located generation fits land-rich campuses that want firm power from day one, as in the recent power-first campuses that build wind, solar, and storage alongside the data hall. Renewables alone rarely serve a critical load every hour, so the credible plans pair contracted clean energy with firming capacity, whether storage, gas, or a grid service, and weigh the renewable resource quality of the site itself.
Fiber and Connectivity: A Front-End Decision
Power dominates the headlines, and connectivity still determines whether a site can serve its intended workload and whether it launches on schedule. Treating fiber as something to resolve after the land is secured is a common and expensive mistake. Confirm at least two physically diverse carrier routes so a single cut cannot isolate the site, score proximity to major Internet Exchange Points and carrier-neutral ecosystems against the workload's latency budget, and for coastal or gateway sites, weigh the subsea cable landings and long-haul routes that can reset the connectivity calculation for an entire region.
Photo: Robert Harker, CC BY-SA 3.0, via Wikimedia Commons.
Latency tolerance varies sharply by workload. Large training clusters can sit far from end users as long as power and cooling scale, while inference and edge deployments depend on sub-10ms windows and have to sit inside a defined latency radius of the population they serve. Pick the workload first, then hold every candidate site to that standard.
Environmental Risk: Climate, Water, and Cooling
For an asset built for near-continuous availability, the physical environment is a first-tier concern. Insurers are pricing it, and stranded-asset risk is now part of serious underwriting. The exposure comes in two forms. Acute hazards like riverine and coastal flooding, wildfire, tropical cyclone wind, and seismic activity threaten the structure directly, while chronic hazards like sustained heat and drought quietly erode cooling efficiency and water availability across the asset's life. Both sit on top of a cross-dependency risk that is easy to miss, since a facility hardened against flooding still fails if the grid, the fuel supply, or the access roads around it go down.
Water has become a siting constraint on par with power, and securing electricity does not guarantee enough cooling water, wastewater capacity, or municipal support. The cooling architecture sets the terms. Evaporative cooling lowers electricity demand while consuming large volumes of water and shifting part of the burden onto municipal systems, whereas dry and adiabatic designs cut water use sharply while raising energy overhead during extreme heat. The direction of travel is clear, with over 40% of new hyperscale capacity announced since 2023 moving toward low- or no-water cooling in response to regulation and community pressure. A 6 MGD request means little if the water authority answers, as a Newton County, Georgia authority recently did, that the water is simply not there.
Assess this at the basin and regional level, well beyond the meter. Recent analysis of nearly 9,000 facilities shows new builds drifting into hotter and drier regions, drawn by cheap power and land even as chronic heat and drought exposure rises, and that trade is defensible only when it is made deliberately, with cooling design and resilience CapEx sized to the hazard profile rather than discovered after commissioning.
The Real Bottleneck: Turning Scattered Data Into Decision-Grade Intelligence
Everything above is well understood across the industry. Good sites still slip away because the data behind each criterion lives in a dozen disconnected places, in formats that resist a fast, defensible comparison.
Interconnection-queue records sit with ISOs and RTOs. Substation and generation data sit in federal filings. Flood layers sit with FEMA, wetlands with the National Wetlands Inventory, broadband with the FCC, and pipelines and fiber corridors in their own systems. A team assembling this by hand spends weeks in GIS, and a broker's shortlist often skips power viability entirely, so the fatal flaw surfaces only after site visits and a drafted LOI.
Three properties separate data you can act on from data that only looks informative. The first is coverage at any coordinate, so a screen works for a raw latitude and longitude anywhere in the U.S. rather than only for known parcels in named markets. The second is provenance and freshness, because interconnection positions change month to month, and every figure that informs a capital decision should carry its source and its vintage so a stale number cannot pass itself off as current. The third is programmatic access, since the site-sourcing that teams now want to automate needs a data backend that software and AI agents can query directly.
This is the problem Mireye Earth was built to address. Point it at any US coordinate and it returns provenance-tagged answers across the factors that decide a site: which ISO or RTO and balancing authority the point falls in and what the nearby interconnection queue looks like, distance to natural gas pipelines and to long-haul fiber and rail corridors, wind and solar resource, coastal distance and surface water and wetlands, elevation and terrain, and flood, heat, drought, and wildfire exposure. Every value cites the source it came from and when it was current.
Because that data is available as a REST API, a natural-language query interface, and an MCP endpoint, it fits the way teams already work. An analyst can ask a plain-language question about a parcel, an underwriting model can pull the same fields programmatically, and an AI site-sourcing agent can run the first-pass screen across hundreds of coordinates before anyone opens a map. The judgment stays with your team. The assembly and the sourcing stop being the bottleneck.
Where to Start
The order of operations is what protects capital. Power and energy strategy first, then connectivity, then the environmental risks that shape resilience and long-term cost, with a clear exit criterion at each stage so a fatal flaw surfaces early rather than after diligence dollars are spent.
- Define load, latency, and timeline thresholds before looking at a single parcel.
- Screen for power and interconnection-queue viability before site visits, so a brokered shortlist cannot quietly waste weeks.
- Re-verify interconnection-queue positions against current data, since they can shift within a quarter.
- Underwrite climate and water over the full life of the asset, and size resilience CapEx to the hazard profile.
The developers who win the next cycle will be the ones who can screen more sites, more rigorously, in less time, on data they can trust. Try Mireye Earth on one of your candidate coordinates right now, no sales call required, at the /compare demo, and see the full power, connectivity, water, and climate picture for a single US coordinate in one place. When you are ready to screen a full shortlist, the same provenance-tagged data runs through our API and an MCP endpoint your team can wire into its own tools.
Frequently asked questions
What is data center site selection?
Data center site selection is the process of screening and ranking candidate land against the physical criteria a facility must clear — power availability and speed-to-power, grid interconnection, fiber connectivity, water for cooling, and climate and hazard exposure — before capital is committed. The aim is to surface any fatal flaw early, while it is still cheap to walk away.
Why is data center site selection now called “grid selection”?
For most of the last decade, fiber density decided where data centers landed. Power deliverability has since become the binding constraint, so the site that wins is the one where firm power can actually be energized on schedule. The industry sums this up as “site selection is now grid selection”: you are really choosing a position on the grid, not just a parcel.
What are the most important data center site selection criteria?
In order of decisiveness: power and energy strategy first (time-to-power at the voltage class you need, interconnection-queue viability, and a firm-power plan), then fiber and connectivity (at least two physically diverse carrier routes and a latency budget that fits the workload), then environmental risk (flood, heat, drought, and wildfire exposure, plus water availability for cooling). Power sits first because a site that fails on power cannot be rescued by strong fiber, cheap land, or a generous incentive.
What is an interconnection queue, and why does it matter for data center siting?
The interconnection queue is the grid operator's backlog of requests to connect new load or generation. Being near a substation or a transmission line is not the same as being able to draw power — the operator still has to study, approve, and build any upgrades. U.S. queues now hold on the order of 2,300 GW, and average waits have climbed from under 2 years in 2008 to roughly 5 years today, past 9 years in California, so queue position is often the real gate on a project's timeline.
How long does it take to get power to a new data center?
In power-constrained markets, time-to-power commonly runs a baseline of 24 to 36 months before the first server is energized, and longer where interconnection queues are congested. That timeline is why developers are shifting from Tier-1 hubs like Northern Virginia to Tier-2 markets such as Columbus, Des Moines, and San Antonio, where power can arrive 12 to 24 months faster and land runs sharply cheaper.
How do you screen land for a data center quickly?
Define your load, latency, and timeline thresholds first, then run power and interconnection-queue viability before any site visit so a brokered shortlist cannot quietly waste weeks on non-viable land. Mireye Earth turns any US coordinate into provenance-tagged data on grid and interconnection, gas pipelines, fiber and rail corridors, water and wetlands, terrain, and flood, heat, drought, and wildfire exposure — available as a REST API, a natural-language query, and an MCP endpoint — so an analyst or an AI agent can run the first-pass screen across hundreds of coordinates before anyone opens a map.
More from the Mireye blog