Where should you deploy your IT infrastructure? Data center selection

AI-driven demand is reshaping not just how data centers are designed, but where they are built. Training clusters are moving to remote locations with abundant, low-cost power. Inference hubs need to be close to people, inside existing cloud zones in major metros.

Meanwhile, enterprise and cloud operators are navigating supply constraints, soaring power costs, and growing regulatory complexity. Location decisions that once turned on land cost and connectivity now involve grid capacity, climate resilience, data sovereignty regulations, and long-term energy pricing.

We look at six important factors in site selection today and explain why these factors may vary depending on whether you're a hyperscaler, a neo cloud, or an enterprise scaling its infrastructure.

Power availability and grid capacity

Power is the single most important constraint in data center site selection today. According to McKinsey, avoiding a global capacity deficit would require building at least twice the data center capacity constructed since 2000 in less than a quarter of the time.

That pressure flows directly onto the grid — and it is compounded by the density of AI workloads themselves. Traditional data centers typically run 3 to 12 kW per rack. AI environments routinely reach 100 kW per rack, creating a sustained high load that conventional power infrastructure was never designed to support.

When evaluating a location for power, operators need to assess three things: available or planned capacity, power redundancy, including dual feeds and backup generation, and grid stability, since regions with aging infrastructure or frequent curtailment carry real operational risk.

For AI training workloads, power availability at scale often trumps proximity. Several of the largest AI training campuses now under construction in Texas, Louisiana, and Indiana were sited primarily for grid access rather than proximity to customers. These remote locations were chosen because of what the grid could deliver at gigawatt scale.

Inference is different: those facilities must be in locations that are already power-dense and grid-connected, sitting inside or adjacent to existing cloud availability zones, because latency to end users matters.

Iron Mountain Data Centers operates more than 30+ data centers across three continents, from Singapore to Virginia to Amsterdam, chosen for grid access and long-term scalability across both Tier 1 and emerging Tier 2 hubs.

Energy pricing and long-term cost predictability

Energy is typically the highest operational cost in a data center, and it varies significantly by location. State and national tax incentives, local utility structures, and the cost of renewable energy in a given market can mean the difference between a profitable deployment and an unsustainable one over a 10- to 15-year facility lifecycle.

Three energy cost factors shape site selection:

• Base power price: kilowatt-hour rates vary dramatically by region, with hydro-, wind-, and solar-rich markets typically offering lower baseload costs.
• Power Usage Effectiveness (PUE): cooler climates support lower PUE naturally; hot markets like Phoenix require active cooling investment. A low design PUE can offset higher power costs over time.
• Renewable availability: operators with RE100 or ESG targets need a colocation partner who can credibly match consumption with renewables and pass that benefit into their sustainability reporting.

Iron Mountain Data Centers powers all facilities with 100% matched renewable energy. Our Green Power Pass (GPP) lets customers count their Iron Mountain power consumption as green energy in CDP, RE100, GRI, and other sustainability frameworks.

Cheaper power is not always better. A low-cost market with an unreliable grid, poor redundancy, or limited renewable supply can create costs and risks that don't show up in the headline rate.

Connectivity and network ecosystems

For AI inference in particular, connectivity is a hard performance constraint. Most production AI applications target sub-50 millisecond response times, which means inference hubs need to sit inside or immediately adjacent to the network ecosystems that serve end users and cloud platforms. A site with great power and land but thin bandwidth options is still the wrong site.

The right ecosystem covers IP transit providers for redundancy, direct cloud on-ramps to AWS, Azure, and Google Cloud, local peering exchanges (IXPs) for low-latency bandwidth aggregation, and physical and virtual cross-connects between colocation customers on the same campus.

Iron Mountain's global network service portfolio enables customers to connect to the carriers, clouds, and partners of their choice. Our Madrid campus, for instance, sits adjacent to the Barcelona-Madrid fiber backbone with direct access to Espanix and DE-CIX peering exchanges, each handling volumes exceeding 1 Tbps. For more on how connectivity fits into a global infrastructure strategy, see our e-book: How to Build a Global Data Center Strategy.

Land availability and expansion potential

Supply constraints are no longer hypothetical. Vacancy rates in Northern Virginia, the world's largest data center market, fell below 1% in 2024.

According to JLL's North America Data Center Report, more than 35 GW of data center capacity is currently under construction across North America — and 92% of it is already pre-committed. Space is being leased two to three years in advance, particularly for larger dedicated builds. The same pattern is playing out in London, Frankfurt, Singapore, and Tokyo.

Operators evaluating sites need to think not just about current requirements, but about where demand will be in five to ten years.

That means looking for:

• Campus-scale land parcels with room for multiple buildings, phased development, and shared infrastructure such as substations and cooling systems.
• Clear zoning and permitting pathways. Markets with predictable processes allow faster time-to-power. Markets with overlapping jurisdictions, environmental objections, or community opposition significantly slow delivery.
• Adjacency to established markets. Spillover markets near saturated Tier 1 hubs often offer land at lower cost with existing connectivity infrastructure already in place.

Northern Virginia illustrates the land challenge well. Development has expanded well beyond Loudoun County's historic data center corridor, pushing south along the I-95 corridor toward Richmond as land and power constraints have tightened.

Iron Mountain Data Centers has made two significant investments in response: a 40-acre campus in Manassas, adding 150 MW adjacent to an existing 250 MW IMDC footprint, and a 66-acre campus in Richmond delivering over 200 MW of AI-ready infrastructure across four facilities. The first Richmond phase is scheduled for Q4 2027.

In Europe, Madrid shows how Tier 2 markets absorb overflow demand. Structure Research forecasts the market growing from 228 MW in 2025 to 1.18 GW by 2030, driven by land availability, green energy, and its role as a southern European connectivity hub. IMDC's Madrid campus, the largest in Spain at 79 MW of planned capacity, was recently put forward by the Spanish government for an EU-funded AI gigafactory initiative.

Climate, environmental risk, and resilience

Site selection decisions made today will shape operational risk for 20 to 30 years. Over that timeframe, climate and environmental risk assessment is simply part of doing the analysis properly.

Key risk categories include:

• Flood risk, both acute (100-year flood events) and chronic (rising groundwater, storm surge).
• Seismic risk: earthquake-prone zones require more expensive structural engineering and carry higher insurance costs.
• Extreme heat: markets like Phoenix require liquid cooling investments that northern European markets do not, and operating costs diverge accordingly.
• Water availability: water-cooled facilities in drought-prone markets face growing regulatory risk. Evaluate water use effectiveness (WUE) alongside PUE.
• Political stability: geopolitical exposure requires contingency planning that adds cost and complexity.

Iron Mountain evaluates climate and environmental risk systematically across our global portfolio. All new data center builds are certified to the BREEAM sustainable construction standard, covering land and water use, energy, ecology, transport, community impact, and recycling. The most recently assessed elements of our Madrid campus achieved an Outstanding BREEAM rating of 93.8%, the highest available.

Regulatory, tax, and incentive landscape

The regulatory environment around data centers is moving in multiple directions at once. Some markets are competing for investment with tax incentives, streamlined permitting, and direct government support. Others are tightening energy regulations or introducing data sovereignty rules that restrict where certain workloads can be hosted. Operators need to evaluate:

• Tax and incentive structures: property tax abatements, sales tax exemptions, and direct investment incentives can materially affect total cost of ownership. The gap between favorable and unfavorable markets can run to tens of millions of dollars over a facility's lifetime.
• Data sovereignty requirements: GDPR, India's Digital Personal Data Protection Act, and comparable frameworks in dozens of markets determine where data can be stored and processed, and increasingly dictate which markets are viable for specific workloads.
• Energy regulations: some markets limit fossil-fueled generation or mandate water use disclosures and efficiency benchmarks.
• Permitting timelines: a market with a 24-month permitting pathway changes the economics of any time-sensitive deployment.

Iron Mountain Data Centers maintains a comprehensive compliance portfolio across our global footprint, including ISO 27001, SOC 2 Type II, PCI-DSS, HIPAA, FISMA HIGH alignment, and BREEAM certification. See our full compliance matrix in the How to Build a Global Data Center Strategy e-book.

Building where it matters

We know that each organization arrives at this decision with different priorities. Hyperscalers need gigawatts of power and multi-hundred-acre land parcels with stable long-term regulatory environments. Neo clouds need capacity available now, carrier-neutral infrastructure, and the dense connectivity ecosystems that support latency-sensitive AI workloads. Enterprises need consistent compliance standards and security protocols across every location, alongside the sustainability credentials that matter for regulatory reporting and corporate ESG commitments. The criteria are not the same, and the right location is not the same either.

Iron Mountain Data Centers has built its 1.4 GW global portfolio to meet those different needs, with facilities in 11 North American markets, three core European hubs, and six Indian markets spanning Mumbai, Chennai, Delhi, Bangalore, Hyderabad, and Pune.

Every location is powered by 100% matched renewable energy and certified to consistent compliance and security standards. Whatever your site selection requirements look like, we can work with you to find the right locations. To explore our full global footprint, download our "How to Build a Global Data Center Strategy" e-book or contact our team.

Frequently asked questions

Is cheaper power always better for data center site selection?

Not necessarily. A low kilowatt-hour rate doesn't mean much if grid reliability is poor, available capacity is constrained, or renewable energy isn't available for sustainability reporting. Operators should weigh energy cost alongside PUE, grid redundancy, water availability, and total infrastructure cost over a 15- to 20-year lifecycle. A stable, renewable-rich market with favorable tax incentives often produces better long-term economics than a cheap-power market with reliability problems.

Why are data centers being built in hot climates like Phoenix or Texas if cooling is more expensive?

Power availability and land cost are usually the deciding factors. Markets like Phoenix and Abilene, Texas, have access to large, contiguous blocks of grid capacity and available land that are increasingly scarce in established Tier 1 markets. For AI training workloads, power scale matters more than ambient temperature, and modern liquid cooling handles high-density AI rack environments effectively regardless of climate.

What is the difference between a Tier 1 and Tier 2 data center market?

Tier 1 markets, including Northern Virginia, London, Frankfurt, Amsterdam, Singapore, and Tokyo, are established hubs with deep connectivity ecosystems, strong cloud presence, and large concentrations of existing infrastructure. They command premium colocation pricing and are often supply-constrained. Tier 2 markets, such as Richmond, Madrid, Manchester, Johor Bahru, and Mumbai, offer lower land and power costs, expanding connectivity infrastructure, and greater capacity. As Tier 1 markets saturate, Tier 2 hubs are absorbing spillover demand fast.

How does data sovereignty affect where enterprises can build or colocate?

Data sovereignty regulations, including the EU's GDPR, India's Digital Personal Data Protection Act, and similar frameworks in dozens of markets, impose restrictions on where certain categories of data can be stored, processed, or transferred. For enterprises operating across multiple markets, these regulations effectively determine which locations are viable for specific workloads. Working with a colocation provider that already has compliance infrastructure in place across multiple jurisdictions considerably reduces the operational complexity of multi-region deployments.