Data Center Site Selection for Corporate Real Estate Teams

Few corporate real estate (CRE) teams are asked to site a hyperscale campus. The usual brief is narrower: replace an aging on-premises data center, add a disaster recovery site, consolidate several server rooms, or secure colocation capacity for a growing private cloud or AI cluster. The decision is shared with IT, security, finance and procurement, and the CRE team’s contribution is the part IT departments rarely do well: market knowledge, lease and purchase negotiation, and site diligence.

The mix of venues has shifted. Uptime Institute estimated in 2025 that about 45% of corporate workloads ran on premises, down from 58% five years earlier, and cautioned that the figures are an approximate trend.⁶ Its 2026 survey found third-party venues (colocation and cloud) holding 46% of enterprise workloads against 44% in company-owned facilities, and it projects 48% versus 42% by 2028.¹ Uptime also notes that security and regulatory requirements keep many workloads on premises, so most enterprises end up hybrid.¹ For the full build, lease or cloud decision, see build vs. lease; for the IT-led view, see enterprise data center site selection.

The most common failure in an enterprise search is starting with buildings instead of requirements. Before contacting brokers or providers, CRE and IT should agree on a short written specification that every option will be judged against. It does not need to be long, but each line should have an owner and a number.

Density deserves care because it drives cost and narrows the field. Uptime’s 2025 survey of 1,677 respondents found average rack densities rising, with more racks in the 10–30 kW range but few facilities above 30 kW.⁷ In its 2026 survey, about 69% of respondents said their most common rack draws less than 10 kW, 20% said 10–19 kW, 8% said 20–49 kW and 3% said 50 kW or more.⁴ A team planning a GPU cluster belongs in that last group and should say so in writing, because many existing colocation halls cannot cool those loads with air; see rack density trends.

Enterprise buyers compete for the same buildings as cloud and AI tenants, and in 2026 that competition sets the timetable. CBRE’s first-half 2026 report on eight primary North American markets counted 7,481 MW under construction, a record, but more than 80% of it was already preleased, leaving less than 1,500 MW available to pre-lease.³ Vacancy held at 1.4%, and in Northern Virginia it fell to 0.2%.³

Three consequences follow for a CRE team. First, start earlier than a typical office search: capacity that will be delivered in two years may already be committed. Second, expect pricing to be quoted per kilowatt of committed power rather than per square foot; CBRE reported asking rents rising for every major deployment size, led by an 8.3% increase for users seeking 3–10 MW.³ Third, consider secondary and emerging markets, where an enterprise requirement of a few megawatts is a meaningful tenant. Our guides to vacancy and pricing and colocation site selection cover those markets.

An enterprise data center is usually tied to the business, not just to cheap power. That makes distance, hazard exposure and separation between sites the location questions that matter most.

Distance and latency

Light in optical fiber adds roughly 5 microseconds of delay per kilometer each way. Cisco’s design guidance notes that synchronous replication, which needs two round trips per write, adds about 1 millisecond at 50 km.⁵ Applications that write synchronously between a primary and a secondary site therefore limit how far apart the two can be, while asynchronous replication allows more distance at the cost of possible data loss. Measured route distance matters more than straight-line distance; see latency requirements by workload.

Separation and hazards

Federal security guidance treats separation as a requirement. NIST SP 800-53, the control catalog behind many federal and corporate programs, includes control CP-7 (Alternate Processing Site) with an enhancement for separation from the primary site.⁸ Banks are examined against the FFIEC Business Continuity Management booklet, which replaced the 2015 continuity planning booklet in 2019.⁹ For a first pass at shared regional risk, FEMA’s National Risk Index maps 18 natural hazards at county and census tract level, though FEMA notes it reflects average past conditions and does not predict the future.¹⁰ Two sites far apart can still share a hazard, a grid or a fiber route, so check those too; see natural hazard risk.

A request for proposal turns the requirements document into comparable offers. Most enterprise searches move through the same stages, though they compress when capacity is scarce.

Ask every provider the same questions in the same format, so answers can be scored side by side:

• Committed power: kW available on the contract date and on what schedule, and whether that capacity is energized today or depends on a future utility upgrade.
• Redundancy: the design topology for power and cooling, and whether any Uptime Tier claim is a certification of design documents, the constructed facility or neither.²
• Expansion: a right of first refusal or reserved capacity, with price and timing.
• Compliance: current third-party audit reports and the physical security controls your auditors will test.
• Network: carriers on site, diverse building entrances and cross-connect pricing.
• Contract terms: service level credits, maintenance notice, pass-through of power costs and exit assistance.

Score in two passes. The first is pass or fail: does the site meet the minimum load, density, redundancy, latency and compliance lines? Anything that fails a gate drops out regardless of price. The second pass weights the survivors on cost, expansion, operator quality and risk; site scoring and weighting methods explains how to set and test the weights.

Tier language is where offers are least comparable. Uptime’s Tier III means planned maintenance or replacement of any capacity component or distribution path can happen without shutting down IT, though the site remains exposed to equipment failure or operator error; Tier IV means a single equipment failure or path interruption will not affect operations.² Tiers I and II require site-wide shutdowns for maintenance.² A provider may describe a hall as “Tier III design” without an Uptime certification, or cite another rating scheme. Ask which document backs the claim, then see tier levels and redundancy.

Some enterprises still build, usually for large, stable loads, strict control requirements or locations where no suitable colocation exists. Owning shifts the hard questions to the land: whether the utility can deliver the load and when, whether zoning allows the use, and how much of the parcel is buildable after flood, wetland and setback constraints. Those questions take months to answer and should run in parallel with design, not after it.

A practical sequence for a CRE team weighing a build:

1. 01Fix the load and ramp in writing, then ask the utility for a preliminary view of service and timing.
2. 02Screen candidate parcels for power, fiber, flood, wetlands and zoning before signing an option.
3. 03Compare the owned option against a colocation offer on the same load, term and redundancy.
4. 04Commission a feasibility study on the finalist before committing capital.

BlackForge screens land for power, water, fiber, terrain, flood and zoning; if you have candidate parcels, you can get a site reviewed.