Asset Pillar · Data Center

    Data center feasibility study.

    Lender-grade feasibility analysis for data center development and acquisition — colocation, hyperscale, edge, AI/GPU clusters, and power-and-site selection. Data center has become the most institutionally-credible asset class in commercial real estate, with AI training cluster demand exceeding supply and multi-billion-dollar projects routinely closing across debt fund, life-co, CMBS SASB, and project-finance capital sources.

    Debt fund · Life-co · CMBS SASB · Project finance · Hyperscaler tenant lease · $8M-$45M per MW capital cost band

    Market Positioning

    Data center as institutional darling — the AI tailwind.

    Three structural shifts since 2023 made data center the most aggressively-financed asset class in commercial real estate. The volume below illustrates the scale.

    $200B+

    Hyperscaler 2025 Capex

    Microsoft, Google, Amazon, Meta, Oracle combined data center capex. Up sharply from prior cycles.

    100MW+

    AI Training Cluster Minimum

    Modern AI cluster deployments. Traditional colocation by contrast runs 10-30MW campus scale.

    Under 1.5

    PUE Target For Institutional Deals

    Power Usage Effectiveness. Hyperscale targets approach 1.1-1.2; older colocation runs 1.5-1.8.

    The post-2022 shift toward generative AI restructured data center demand at the foundation. Training-cluster GPU deployments require campus-scale 100MW-plus facilities, often co-located with major utility transmission lines and water resources for liquid cooling. Inference workloads follow conventional colocation density patterns. Traditional cloud and enterprise colocation, the historical core of the asset class, runs at 10-30MW campus scale. The three demand layers compete for the same constrained inputs: power, water, fiber connectivity, and entitled land in tier-1 markets.

    Hyperscaler capital expenditure validates the institutional thesis. Microsoft, Google, Amazon Web Services, Meta, and Oracle collectively committed over $200 billion in 2025 data center capex, with similar or larger 2026 commitments announced. Capital flows: debt funds and bridge lenders for ground-up development; life-insurance companies and CMBS SASB for stabilized credit-tenant deals; project-finance structures for hyperscale campus phasing. Healthcare REITs and traditional commercial REITs do not compete in this asset class; specialty data center REITs (Digital Realty, Equinix, Iron Mountain Data Centers) and private vehicles dominate the institutional ownership universe.

    The infrastructure constraint is real and binding. Power availability and grid capacity, not real estate per se, determines where new data center capacity can land. Site selection follows utility transmission, not traditional CRE locational logic. The bankable framework's data center scope is built around the infrastructure stack — power, fiber, PUE, water — alongside the conventional CRE analytical lenses.

    Lender Matrix

    Data center lending by capital source and deal type.

    Each capital source serves a different point in the data center lifecycle. Development risk goes to debt funds and project finance; stabilized credit-tenant deals attract life-co and CMBS SASB.

    Deal TypeDebt Fund / BridgeLife-CoCMBS SASBProject FinanceBank Construction
    Hyperscale ground-up (single tenant)
    Strong
    development
    pre-stabilization
    Strong
    phased campus
    Good
    construction
    Hyperscale stabilized (long-term lease)
    Limited
    refinance only
    Strong
    preferred
    Strong
    SASB-friendly
    Good
    refinance
    AI / GPU cluster build (100MW+)
    Strong
    specialty fund
    specialty risk
    pre-stabilization
    Strong
    institutional
    Good
    construction with takeout
    Colocation ground-up (multi-tenant)
    Strong
    development
    Limited
    lease-up risk
    Limited
    pre-stabilization
    Fair
    smaller scale
    Good
    construction
    Colocation stabilized
    Fair
    refinance
    Good
    institutional
    Good
    pool-friendly
    Good
    refinance
    Edge data center (sub-10MW)
    Fair
    smaller mandate
    Limited
    smaller mandate
    Limited
    pool-size
    Limited
    smaller scale
    Good
    regional bank
    Powered shell (no tenant fit)
    Strong
    dev land
    Limited
    specialty
    unstabilized
    Fair
    phased
    Good
    construction
    Conversion / repositioning
    Strong
    value-add
    specialty risk
    Limited
    specialty risk
    Limited
    Fair
    case-by-case

    Cell ratings reflect typical 2026 underwriting posture. Hyperscaler-tenanted deals carry distinct credit underwriting that compresses pricing across all relevant lender types. Project finance structures typically require commitment from a hyperscaler tenant before debt commitment.

    Capital Cost

    Capital cost per MW by data center type.

    Data center capital cost is benchmarked per megawatt of IT load capacity, not per square foot. Power density and cooling configuration drive most of the variance.

    AI / GPU cluster (training)

    SHELL-AND-CORE: $12M-$18M per MW

    FULLY FITTED: $30M-$45M+ per MW

    Liquid cooling infrastructure, high-density power distribution, redundant transmission. Fully-fitted cost includes GPU rack-and-stack at hyperscaler standards. Cost runs 1.5-2x traditional colocation per MW.

    Hyperscale (cloud / enterprise)

    SHELL-AND-CORE: $9M-$13M per MW

    FULLY FITTED: $25M-$35M per MW

    Single-tenant or hyperscaler-leased. Standard rack densities, conventional cooling, 100MW+ campus scale typical. Capital cost has trended upward with land and power constraints.

    Wholesale colocation

    SHELL-AND-CORE: $8M-$12M per MW

    FULLY FITTED: $22M-$32M per MW

    Multi-tenant wholesale, larger suite configurations (1MW-5MW per tenant). Mix of credit and non-credit tenants. Stabilized lease-up takes 18-36 months typical.

    Retail colocation

    SHELL-AND-CORE: $9M-$13M per MW

    FULLY FITTED: $24M-$34M per MW

    Sub-1MW tenant footprints. Equinix and Digital Realty dominate the institutional retail colo market. Cost premium reflects more granular power distribution and operational overhead.

    Edge data center (sub-10MW)

    SHELL-AND-CORE: $10M-$15M per MW

    FULLY FITTED: $25M-$38M per MW

    Small-footprint, latency-driven deployments adjacent to population centers. 5G mobile edge, content delivery, IoT processing. Higher per-MW cost reflects smaller scale economics.

    Powered shell / pad-ready

    SHELL-AND-CORE: $5M-$9M per MW

    FULLY FITTED: n/a (sold or leased to fit-out tenant)

    Land entitled, power secured, building shell complete. Tenant or buyer completes fit-out. Lower base cost; resale or lease-up timing drives the deal.

    Bands reflect 2026 hard cost in tier-1 US markets (Northern Virginia, Dallas, Phoenix, Silicon Valley, Hillsboro). Tier-2 markets (Columbus, Reno, San Antonio, Salt Lake City) trend 5-15 percent below the band. Premium markets with severe land or power constraints can exceed the band's top end by 20 percent or more.

    Infrastructure Constraints

    Power, fiber, PUE, and water — the four binding constraints.

    Data center feasibility analysis routes through four infrastructure constraints. Each constraint can be the deal-killer; bankable scope quantifies all four.

    01

    Power availability and grid capacity

    Power is the foundation. Site selection in 2026 follows utility transmission corridor more than traditional CRE locational logic. Substation capacity, transmission line proximity, and time-to-energization timeline (often 24-48 months in tier-1 markets) determine project viability. Power cost in $/kWh varies 3-5x across markets — Pacific Northwest hydro economics differ fundamentally from Northern Virginia coal-and-gas grid economics.

    02

    Fiber and connectivity infrastructure

    Fiber availability and carrier density drive operational viability. Carrier-neutral cross-connect points within the building and submarket multi-carrier presence affect tenant decision-making. Latency to major peering points (Ashburn, Dallas, Silicon Valley) determines edge versus core positioning. Long-haul fiber to hyperscaler peering hubs is essential infrastructure.

    03

    PUE benchmarking

    Power Usage Effectiveness measures total facility power divided by IT load power. PUE of 1.0 means zero overhead; 1.5 means 50 percent overhead. Modern hyperscale runs 1.1-1.25; older colocation runs 1.5-1.8. Lender expectations cluster around sub-1.5 for institutional deals; AI cluster underwriting expects sub-1.3 with liquid cooling configurations.

    04

    Water and cooling analysis

    Liquid cooling for AI clusters and high-density configurations requires substantial water access. Water rights, withdrawal permits, and ongoing supply economics constrain site selection. Climate suitability — cooler ambient temperature reduces cooling load — drives Pacific Northwest, Nordic, and high-elevation siting. Markets with water stress (Phoenix, parts of California) face increasing regulatory friction.

    Market Tiers

    Primary markets and emerging tier-2 markets.

    Data center capacity concentrates in a handful of markets historically. Tier-2 markets are absorbing rapidly under power and land pressure in tier-1.

    MarketTierInventory (MW)Power EconomicsDominant Use
    Northern Virginia (Ashburn)Tier 13,000+ MWTight; rising ratesHyperscale + colocation
    Dallas–Fort WorthTier 1800+ MWModerateHyperscale + AI clusters
    PhoenixTier 1700+ MWTight; water concernsHyperscale + AI clusters
    Silicon ValleyTier 1500+ MWVery tight; constrainedEnterprise + AI inference
    ChicagoTier 1500+ MWModerateHyperscale + colocation
    AtlantaTier 1400+ MWModerateHyperscale + colocation
    Hillsboro / Pacific NWTier 1400+ MWHydro; cool climateHyperscale + AI clusters
    Columbus, OHTier 2 (emerging)250+ MW; risingModerateHyperscale (Intel adjacency)
    Reno / Northern NevadaTier 2 (emerging)200+ MW; risingCool climate; water concernsHyperscale + AI
    San AntonioTier 2 (emerging)150+ MW; risingModerate; growingHyperscale
    Salt Lake CityTier 2 (emerging)150+ MW; risingCool climate; growingEnterprise + hyperscale

    Inventory figures reflect approximate operational capacity as of 2025. Pipeline development in tier-2 markets is rapidly closing the gap with tier-1 incumbents. Microsoft, Google, Amazon, Meta, and Oracle development announcements drive most of the tier-2 absorption.

    AI vs Traditional

    AI/GPU cluster versus traditional colocation — underwriting differences.

    AI training clusters and traditional colocation operate at fundamentally different scales and demand fundamentally different feasibility methodology. The matrix below covers the practical differences.

    DimensionAI / GPU ClusterTraditional Colocation
    Typical campus scale100MW-500MW+10MW-30MW
    Tenant profileSingle hyperscaler training tenantMulti-tenant colo or single enterprise
    Cooling configurationLiquid cooling requiredAir cooling typical
    Power density per cabinet50-100kW+ per rack5-15kW per rack
    Fiber latency requirementLess critical (training is remote-tolerable)Critical for inference and enterprise
    Site selection priorityPower + water + climate + transmissionFiber + tenant catchment + latency
    Capital cost per MW (fully fitted)$30M-$45M+$22M-$34M
    Lease term typical10-20 years (hyperscaler)3-7 years (colo) / 10+ years (single tenant)
    Capital source typicalProject finance + debt fund / hyperscaler balance sheetLife-co + CMBS + debt fund
    Feasibility scope emphasisPower + transmission + tenant creditFiber + tenant mix + lease comp

    Many modern campuses combine both — a 200MW campus with 100MW dedicated to a hyperscaler AI training tenant and 100MW reserved for traditional colocation use. Hybrid campus underwriting addresses both lenses.

    Capital By Stage

    Capital sources track the deal stage.

    Data center capital flows through distinct sources at distinct deal stages. Sponsors aligning capital strategy to deal stage avoid friction at lender introduction; sponsors who approach a stabilized-asset lender with a development deal — or vice versa — burn time and credibility.

    Development and construction. Debt funds (Blackstone Real Estate Debt Strategies, Brookfield, KKR Real Estate Finance, Starwood Property Trust, Ares, Mesa West) and bank construction lenders fund ground-up data center development. Project finance structures with hyperscaler tenant commitments enable single-tenant campus phasing at $200M-$1B+ scale. Bridge structures typically run 75-80 percent loan-to-cost, floating rate, 2-3 year terms with extension options. Construction lender takeout assumptions drive much of the underwriting — life-co or CMBS SASB exit at stabilization is the most common path.

    Stabilized credit-tenant. Life-insurance companies aggressively pursue stabilized hyperscaler-tenanted data centers. Long-term hyperscaler leases (10-20 years typical) align with life-co's balance-sheet hold preference. PGIM, MetLife, Northwestern Mutual, Pacific Life, and TIAA-Nuveen have all built data center allocations. CMBS SASB transactions securitize trophy data center portfolios at $500M-$2B+ scale, with rating agency methodology drawn from KBRA and S&P published criteria for data center evaluation.

    Project finance and hyperscaler balance sheet. Hyperscalers (Microsoft, Google, Amazon, Meta, Oracle) finance much of their own data center capex on balance sheet, in joint venture structures with developers, or through project-finance vehicles that combine debt, tax equity (for renewable-power-paired campuses), and sponsor equity. Project-finance feasibility scope is denser than typical CRE feasibility — power availability, grid energization timing, tenant credit, and offtake economics all factor into the analysis.

    Conversion and repositioning. A growing subset of data center deals involves converting industrial, office, or specialty properties into data center use. Powered land with existing utility infrastructure, brownfield power generation sites, and decommissioned manufacturing facilities frequently re-enter the market as data center conversions. The bankable framework's data center scope addresses conversion economics, including stranded asset risk if conversion does not stabilize as planned.

    Methodology Applied

    How the bankable framework adapts to data center scope.

    Six methodology components specific to data center feasibility analysis. Each adapts the bankable framework's structural approach to the asset class's distinctive analytical demands.

    Power availability and cost analysis

    Substation capacity, transmission line proximity, time-to-energization, and $/kWh cost modeling. Power constraint analysis is the foundation; deals that fail the power test fail at feasibility regardless of other characteristics.

    Fiber and connectivity assessment

    Carrier-neutral cross-connect availability, multi-carrier presence, latency to major peering hubs, and long-haul fiber capacity. Fiber methodology adapts to asset use — edge demands tighter latency, hyperscale demands carrier diversity, AI training tolerates higher latency.

    PUE benchmarking and cooling design

    Power Usage Effectiveness modeling at design and projected operational levels. Cooling configuration analysis (air, evaporative, liquid, immersion). Lender expectations of sub-1.5 PUE for institutional deals; sub-1.3 for AI cluster.

    Water and climate analysis

    Water rights documentation, withdrawal permits, ongoing supply economics. Climate suitability for ambient cooling efficiency. Regulatory environment for water-intensive cooling configurations.

    Tenant demand by market tier

    Hyperscaler announcement tracking, colocation absorption, AI cluster development pipeline, edge deployment patterns. Demand modeling varies sharply across tier-1, tier-2, and emerging markets.

    Capital cost per MW and exit assumption

    Capital cost benchmarked per MW (not per SF) at shell-and-core and fully-fitted basis. Exit cap rate calibrated to tenant credit, lease term, market tier, and 2026 dynamics. Project-finance scenarios model offtake economics through deal life.

    FAQ

    Data center feasibility frequently asked questions.

    The structural drivers — generative AI training compute demand, inference workload growth, and enterprise AI adoption — extend through at least the 2026-2028 cycle visible from current hyperscaler capex announcements. Cyclical caution applies to specific sub-markets where pipeline outpaces tenant commitments, but aggregate demand remains constrained by power availability rather than tenant absorption. The bankable framework's data center scope addresses both layers.

    Per megawatt of IT load capacity, not per square foot. The convention reflects the binding constraint — power, not real estate. Capital cost runs $8M-$15M per MW for shell-and-core construction in tier-1 markets and $25M-$45M per MW fully fitted depending on cooling configuration, redundancy specification, and tenant fit-out. AI cluster builds run at the top of the band or above.

    24-48 months from utility application to commercial operation in 2026, with substantial market variability. Northern Virginia and Phoenix face the tightest constraints; emerging tier-2 markets often offer 12-24 month timelines as competitive advantage. The bankable framework's power analysis quantifies time-to-energization explicitly because it can determine project viability.

    Yes. Microsoft, Google, Amazon, Meta, and Oracle all carry investment-grade credit profiles that translate cleanly to credit-tenant net lease underwriting. The structural difference from traditional CTNL is lease term (10-20 years typical for hyperscaler) and rent escalator structure (often CPI-linked). Lender pricing tightens substantially for hyperscaler-tenanted stabilized data centers.

    $18,000-$30,000 for tier-2 deal scope (single-program, conventional colocation or stabilized hyperscale). Tier-3 scope (AI cluster, hyperscale ground-up, project-finance structures, complex power-and-water analysis) runs $25,000-$45,000+ depending on infrastructure complexity. Engagement letter sets fixed fee.

    Conversion scope addresses three additional analytical layers beyond ground-up: existing infrastructure capacity (power, water, fiber), conversion cost and timeline, and stranded asset risk if conversion fails to stabilize. Powered shell brownfield with existing utility infrastructure runs more favorable conversion economics than pure structural retrofit.

    Increasingly yes. Columbus (driven by Intel adjacency), Reno, San Antonio, and Salt Lake City have all attracted hyperscaler commitments and increasingly compete on power availability, time-to-energization, and total project economics. Tier-1 incumbency advantages persist for fiber-density-dependent uses (financial services, content delivery, latency-critical inference) but tier-2 advantages emerge for hyperscale and AI cluster deployments.

    Tier 2 data center deals (stabilized colocation, single-program scope) typically run 25-35 business days. Tier 3 deals (AI cluster, hyperscale ground-up, project-finance structures, conversion) run 35-55 business days. Rush turnaround is rarely available in this asset class because power-and-infrastructure analysis requires utility correspondence and substantial primary research.

    Get a data center feasibility study.

    Colocation, hyperscale, edge, AI/GPU, or pre-development power-and-site selection. Single-program or cross-program scope. 30-minute scoping call. Fixed-fee proposal within 24 hours.

    Or read the debt fund deep-dive · Life-co deep-dive · CMBS SASB deep-dive

    Where we prepare data center feasibility studies

    State-specific data center feasibility studies are available in the markets listed below.

    Browse all state feasibility study hubs →