International Site Selection Criteria for Hyperscale Data Center Expansion

International hyperscale site selection has stopped being a real estate exercise. It is now an infrastructure problem, and the first question is whether the grid can deliver power at the scale AI infrastructure demands. Every other criterion resolves inside that constraint.

How power scarcity replaced real estate as the primary filter in hyperscale site selection

Power availability now determines project success more than capital or construction timelines. The traditional ranking that put fiber proximity, real estate cost, and population density first has been displaced by a single gate: can the site get grid-connected electricity, and can it get it on schedule. A site that clears every other criterion but cannot demonstrate a credible path to interconnection within the project timeline is not viable, regardless of land cost or permitting ease.

The numbers make the shift concrete. Modern hyperscale builds run from 100 MW to 500 MW per campus, and AI campuses are now planned in gigawatts. At that load, the grid interconnection queue defines the schedule, not the land parcel. Northern Virginia, the largest data center market globally, illustrates the problem: wait times for new grid connections can exceed seven years. The bottleneck sits upstream of everything else.

Ramboll's site-selection analysis confirms that reliable and abundant power supply is the backbone of any data center site decision. But it is not enough to confirm present capacity. Scalable access must be evaluated: the site needs a path to more power over time, not just what the substation offers tomorrow. Teams that treat power as one variable among many are now making the same mistake that teams made a decade ago by treating real estate cost as the deciding factor.

What grid interconnection queues and on-site generation strategies mean for site feasibility

Because utility interconnection timelines now routinely exceed project delivery schedules, feasibility assessment must evaluate grid access and on-site generation in parallel, not sequentially. Grid-side evaluation means checking substation proximity, available transmission capacity, and where you really are in the interconnection queue. A substation nearby is not the same as a substation with spare capacity. The question is whether consistent, reliable energy is available at the site, and at what timeline and cost.

On-site generation is now a first-class option, not a backup contingency. Ramboll identifies microgrids, battery energy storage systems, small modular nuclear reactors, fuel cells, and natural gas plants as viable on-site solutions that must be evaluated during site selection. These are not afterthoughts for when the grid fails. They are alternative paths to power that can change a site's feasibility entirely.

The evaluation is a risk-weighting exercise. A site with a shorter grid queue but aging transmission infrastructure may carry more long-term risk than a site with a longer queue but modern substation capacity. Teams that treat the queue position as the single number are missing the structural condition behind it. Site feasibility now means understanding both the queue and the physical health of the infrastructure carrying the power.

How regulatory environments function as site veto mechanisms, not just compliance checklists

Regulatory conditions are not a permitting formality to manage after site selection. In several major markets, they are hard veto conditions that eliminate otherwise viable sites before engineering begins. National-level restrictions can close entire geographies outright.

The Netherlands has banned new hyperscale data centers except in designated areas, citing large power consumption and spatial control. A site that clears power, land, and connectivity criteria is still eliminated if it falls outside those zones. The regulatory environment is not the last check. It is the gate that can render all prior analysis moot.

Efficiency mandates set a technical performance floor that constrains design before a line is drawn. Germany's Energy Efficiency Act requires data centers to use a substantial and rising share of renewables, with targets escalating through 2027. The act permits subsidized renewable sources, but the trajectory is fixed. From January 2025, new EU facilities in cooler climates whose operators have signed the Climate Neutral Data Centre Pact have committed to meet a PUE of 1.3 or less. That threshold directly shapes cooling system design and, by extension, site suitability.

Sub-national pressure is intensifying. In Texas in August 2026, the Governor directed PUCT and ERCOT to gather information on the impacts of data centers on local communities, covering tax benefits, water consumption, and on-site generation for all developers pursuing interconnection. ERCOT separately issued community-impact RFIs to loads of 25 MW or more not already covered by the Batch Zero process for loads 75 MW and above.

Zoning and permitting must be navigated early. Building height, setbacks, air and noise levels all matter. Virginia has enacted data center legislation, section 15.2-2209.5 effective July 1, 2026, requiring mandatory sound studies for large facilities. The law sets no noise limits and includes no public notification requirements, and broader noise abatement and notification bills were proposed but not enacted. Phoenix has adopted specific data center ordinances. Regulatory due diligence must happen at the same stage as power and land assessment. Late discovery of a regulatory barrier is one of the most common sources of costly project delays.

Community acceptance and social license as a practical constraint on site approval timelines

Community opposition can delay or defeat a technically viable site. Increasingly documented as data center NIMBYism, it makes early stakeholder engagement a functional requirement of site assessment, not a public-relations afterthought. Ramboll's site-selection criteria explicitly identify community support as pivotal, noting that data center projects frequently face resistance on environmental grounds.

The sources of community concern cluster around energy consumption, water use, noise, and local economic impact. These are the same dimensions that regulators are now formalizing into reporting requirements, as the Texas example shows. The regulatory and community tracks are converging. A project team that prepares for one while ignoring the other is preparing for half the schedule risk.

Ramboll's experience shows that partnerships with municipalities, particularly those transitioning from industrial economies, can facilitate long-term project success. Transparent communication, public consultations, and community benefit programs are the operational tools. Their value depends on timing. Engagement that follows a permit application is defensive. Engagement that precedes it is strategic. Social license is a schedule variable, and it resolves on the same timeline as grid queues and zoning reviews.

Cooling resources, water access, and climate as site qualifiers that interact with power and regulation

A site's climate and water profile are not independent variables. They directly determine whether a facility can meet efficiency mandates, and at what capital and operating cost. Cooler climates lower cooling costs and enable passive or hybrid cooling strategies that support PUE targets. Warmer regions increase energy expenses and constrain cooling design options. The climate profile changes the engineering brief.

Water availability adds a second dimension. Proximity to chilled water plants and sustainable water sources reduces operating costs and supports long-term continuity, especially in regions with water scarcity or strict water-use policies. Natural disaster risk completes the physical site picture. Ramboll's assessments show some regions are unsuitable for hyperscale data centers due to climate risks. Earthquakes, floods, and storms must be evaluated through formal risk assessment, not ruled out by assumption.

The interaction effect is the key point. A site in a cooler, lower-risk climate with accessible water may be able to meet PUE of 1.3 or less at lower capital cost than a warmer site that requires advanced liquid cooling infrastructure to hit the same threshold. Climate is not just an operating-cost variable. It is a compliance variable, and it connects directly back to the EU PUE mandate established earlier.

Why secondary markets and new geographies are gaining traction as primary markets saturate

The saturation of primary data center markets is a constraint signal, not a demand signal. Power and land exhaustion in legacy hubs is actively redirecting hyperscale investment toward secondary markets and new geographies that can meet the full criteria stack. In Europe, the FLAPD markets, Frankfurt, London, Amsterdam, Paris, and Dublin, hold the largest concentration of existing capacity. A 2025 analysis of European data center site selection dynamics found that approximately 40% of total installed European capacity is concentrated within those markets. Grid congestion in those legacy hubs is documented and driving a clear shift toward secondary markets and the Nordics.

New geographies are emerging on the basis of specific infrastructure advantages. Ramboll's site-selection framework notes that hyperscale and colocation data centers are expanding into secondary and remote areas that offer consistent temperatures, water availability, and renewable energy. These are the criteria that primary markets can no longer reliably provide. The Kazakhstan and India examples are instructive because they show how power infrastructure readiness, not just availability, distinguishes credible new geographies from speculative ones. A market that advertises available power but lacks the transmission and distribution infrastructure to deliver it is not a credible alternative.

The objection is real. Secondary markets and new geographies often lack the fiber density, skilled labor pools, and operational ecosystem of primary hubs. Network connectivity and workforce availability must be evaluated alongside power and regulatory criteria, not after. A site that solves the power problem while ignoring the connectivity and labor gaps has traded one constraint for another.

Network connectivity, latency, and workforce as criteria whose weight varies by workload type

Network connectivity and workforce proximity are not universal hyperscale requirements. Their weight in site evaluation shifts materially depending on whether the facility serves latency-sensitive cloud workloads or large-batch AI training. Teams that treat them as fixed criteria will either over-constrain or under-evaluate their options.

Ramboll's framework notes that for AI data centers, latency is less critical. This allows more flexibility in site location and directly expands the viable geography for AI campuses beyond traditional network hubs. For cloud and colocation workloads, proximity to major fiber routes and carrier hotels remains a hard requirement. Locations near cloud providers and network exchanges directly reduce latency and improve application performance.

Workforce availability is a real constraint that secondary markets must address. Experienced technicians, engineers, and support staff are critical for ongoing operations. Proximity to a strong talent pool simplifies staffing and maintenance planning. A secondary market with strong power and regulatory conditions but thin engineering labor supply introduces operational risk that is not visible in a physical site assessment. The framework must be calibrated to workload type, not applied uniformly.

How the criteria interact in practice and why resolving them before design begins changes project outcomes

The criteria do not resolve in isolation. A decision on power sourcing shapes cooling system design, which affects PUE compliance, which interacts with water policy, which influences community engagement. All of these must be resolved before design can generate useful output. Ramboll's analysis is explicit: neglecting early-stage assessments leads to costly delays and wasted resources. Commercial certainty and long-term efficiency depend on due diligence at the site-selection stage, not the design stage.

The failure mode is well documented. When site criteria are resolved partially or sequentially rather than in parallel, design work begins on assumptions that subsequent site findings invalidate. Power routing changes, cooling technology changes, or a regulatory finding forces a redesign of systems already modeled. Total cost of ownership must be evaluated across the full criteria set. A site with low upfront costs may be more expensive to maintain. The interaction of energy cost, staffing, real estate, and tax environment determines the true TCO, and this analysis requires structured data, not a spreadsheet of independent line items.

Design automation matters here. Once site parameters are established, including power capacity, cooling constraints, regulatory thresholds, and layout boundaries, a connected design workflow should propagate those parameters into data hall layout generation, critical-systems sizing, and documentation without manual re-entry at each stage. The fragmentation that characterizes today's delivery workflows means site decisions made in planning rarely survive intact into design and operations without being manually re-translated at every handoff.

ArchiLabs Studio addresses this directly. By connecting site-resolved parameters to layout generation, critical-systems coordination, and operations-ready handoff in a single structured workflow, it removes the manual re-entry that currently causes site criteria to degrade in fidelity as they pass through power models, cooling schedules, cable routing, and DCIM configuration. The structured data that site selection produces should be the same structured data that reaches the operations team at turnover. Incomplete site resolution before design is the root cause of the rework that costs hyperscale projects the most time.