The U.S. data center market is experiencing a demand acceleration with no historical precedent in commercial real estate. National vacancy in primary markets has reached 1.4% — a figure that is not a market condition but an artifact of pre-commitment: 92% of capacity under construction is already pre-leased before delivery.
The U.S. data center market is experiencing a demand acceleration with no historical precedent in commercial real estate. National vacancy in primary markets has reached 1.4% — a figure that is not a market condition but an artifact of pre-commitment: 92% of capacity under construction is already pre-leased before delivery. The AI infrastructure buildout is compressing the analytical framework for feasibility evaluation to a single gating variable — power availability — against which all other site selection criteria are secondary. Development costs have risen 39% since 2020 to an average of $11.7 million per megawatt, with AI-optimized facilities commanding $20+ million per MW. Lease rates have surged to $195.94/kW/month on average — an 11% CAGR since 2020 — and combined Big Five hyperscaler capital expenditures are forecast to reach $602 billion in 2026, representing a 36% year-over-year increase. For lenders and investors evaluating data center feasibility, the analytical challenge has fundamentally shifted: this is no longer a question of whether demand exists, but whether a specific site can obtain power, in what quantity, and on what timeline.
1. Market Scale and National Demand Dynamics
The U.S. data center market reached $121 billion in 2024 by revenue, with broader investment-inclusive measures placing total market activity at $208 billion.(1,2) U.S. colocation inventory grew to 23.7 GW by year-end 2025, up 33% year-over-year — a supply growth rate that nonetheless failed to satisfy demand.(3) Total U.S. installed capacity across all facility types stands at approximately 57 GW in 2025, projected to reach 110 GW by 2030, implying a near-doubling of national capacity within the decade.(4) The market is growing at 10–11% CAGR by revenue and 17% CAGR by supply in the Americas through 2030.(5)
The vacancy figures define the demand imbalance with unusual precision. Primary U.S. markets registered just 1.4% vacancy at year-end 2025 per CBRE, with JLL reporting 1% for the second consecutive year.(6) A remarkable 92% of capacity under construction is already pre-committed, and 74.3% is preleased as of H1 2025 — with preleasing now extending to capacity scheduled for delivery in 2027 and beyond. These figures represent a supply-demand condition with no analog in conventional commercial real estate underwriting.
2. Geography: Primary Markets and Emerging Corridors
Northern Virginia remains the world's largest data center market, with 4,040 MW of total inventory and 0.76% vacancy — a figure so low that meaningful differentiation between available and committed inventory is functionally irrelevant.(6) Net absorption reached 1,102 MW in 2025, and 2,078 MW is under construction, representing an 80% increase. However, Loudoun County now requires special exception zoning for new data center development, and Dominion Energy faces another 4,000 MW of data center load by 2028 — infrastructure constraints that are reshaping developer site selection strategy in the region.
Atlanta experienced transformational 2024 growth, with absorption of 705.8 MW — 39 times higher than 2023 — and inventory growth of 222% to over 1,000 MW, driven by a committed $11 billion AWS investment in the region. Dallas-Fort Worth absorbed 471 MW in 2025 (up 424 MW year-over-year) and is expected to double in total size by year-end 2026. Phoenix grew 67% in 2024 to 603 MW with 1.7% vacancy. Emerging markets include Columbus, Ohio (where hyperscalers acquired approximately 1,900 acres in 24 months), northern Louisiana (Meta/Blue Owl's $27 billion 'Hyperion' campus scaling to 5 GW), and Wisconsin's $15 billion 'Lighthouse' campus involving Vantage, Oracle, and OpenAI.
| Market | Total Inventory | Vacancy | Key 2024–2025 Activity |
|---|---|---|---|
| Northern Virginia | 4,040 MW | 0.76% | 1,102 MW absorbed; 2,078 MW u/c |
| Atlanta | >1,000 MW (+222%) | ~3% | 705.8 MW absorbed; $11B AWS commitment |
| Dallas-Fort Worth | Doubling by YE 2026 | ~2% | 471 MW absorbed in 2025 (+424 MW YoY) |
| Phoenix | 603 MW (+67% in 2024) | 1.7% | Rapid hyperscale expansion ongoing |
| Charlotte/Raleigh | Emerging | ~5% | 113 MW u/c; 90% preleased |
| Austin/San Antonio | Emerging (fastest CAGR) | ~4% | 464 MW u/c; 6.86% CAGR |
| Columbus, OH | Rapid expansion | <3% | ~1,900 acres acquired by hyperscalers |
3. Power: The Binding Constraint on Feasibility
Power availability has replaced cost of debt as the single most cited investment challenge in data center development, identified by 39% of investors in the CBRE 2025 Investor Survey.(6) The PJM interconnection queue holds 30+ GW of data center requests with wait times exceeding 36 months to five or more years. PJM capacity prices surged from $28.92/MW-day for the 2024/25 delivery year to $329.17/MW-day for 2026/27 — an eleven-fold increase in two delivery years — reflecting the systemic stress that data center load growth is imposing on regional grid infrastructure.(9)
U.S. data centers consumed 183 TWh in 2024, representing over 4% of total U.S. electricity generation, projected to reach 426 TWh by 2030 — a 133% increase.(9) AI workloads are compounding this trajectory at the rack level: traditional data center racks run at 5–10 kW, while AI and GPU racks demand 40–130 kW, with NVIDIA's GB200 NVL72 operating at 120+ kW per rack. Future 2nm chip architectures could require 250 kW per rack — a density that existing grid infrastructure and building mechanical systems are not designed to accommodate. Gartner predicts power shortages will constrain 40% of AI data centers by 2027, a forecast that should be incorporated into any feasibility analysis for AI-optimized facilities with post-2025 delivery assumptions.(10)
In response to these infrastructure constraints, major hyperscalers are making direct investments in power generation. Microsoft signed a 20-year power purchase agreement to reopen Three Mile Island (835 MW nuclear), Amazon committed $500 million to X-energy small modular reactors targeting 320 MW, and Google has contracted for 420 MW of solar in Nebraska alongside investments in Kairos Power SMR technology.
4. Development Costs: The Most Capital-Intensive Asset in CRE
Average development costs across 19 U.S. markets reached $11.7 million per MW in 2025, compared with $7.7 million globally in 2020 — a 52% increase in five years.(7) AI-optimized facilities command $20+ million per MW, reflecting the mechanical, electrical, and structural requirements of high-density computing infrastructure. The cost range by market is significant: San Antonio represents the most cost-efficient primary market at $9.3 million per MW, while Reno reaches $15 million per MW — a 61% premium that reflects land, labor, and utility infrastructure differentials.(7)
On a per-square-foot basis, standard data center construction runs $600–$1,100/SF, with prime markets reaching up to $2,200/SF and a national average that has risen to approximately $1,000/SF — a 50% year-over-year increase that has materially altered the feasibility economics for projects budgeted on pre-2024 cost assumptions.(7) A 700,000 SF / 60 MW facility in Northern Virginia carries an all-in cost of $420–$770 million. MEP systems consume up to 50% of total project budget, and equipment lead times for switchgear, chillers, and uninterruptible power supply systems now run 36–48 weeks, extending development timelines and creating procurement risk for projects dependent on rapid delivery schedules.
Brownfield redevelopment offers a meaningful cost advantage — $7–$8 million per MW, or a 10–15% discount to greenfield — and is increasingly competitive in markets where land with existing power infrastructure is available. Average land parcel size has grown to 224 acres (up 144% since 2022), with emerging and tertiary markets averaging 340 acres, reflecting the expanding geographic footprint required to accommodate hyperscale campus development at the scale now demanded.(7)
| Metric | Cost | Source |
|---|---|---|
| Avg. development cost/MW (19 U.S. markets) | $11.7M/MW | C&W 2025 Development Cost Guide |
| Range by market | $9.3M (San Antonio) to $15M (Reno) | C&W 2025 |
| Global avg. construction cost/MW | $10.7M (2025) vs. $7.7M (2020) | JLL 2026 |
| AI-optimized facilities | $20M+/MW | Industry Consensus 2025 |
| Brownfield redevelopment | $7–$8M/MW (10–15% cost advantage) | Dgtl Infra / QTS |
| Per SF — standard | $600–$1,100/SF | Dgtl Infra, 2025 |
| Per SF — prime markets | Up to $2,200/SF | BlueCap Economic Advisors |
| National average per SF | ~$1,000/SF (up 50% YoY) | Programs.com, 2025 |
| Powered shell only | $105–$235/SF (17–21% of total) | Dgtl Infra, 2025 |
5. Lease Rates and Investment Returns
Wholesale colocation asking rates averaged $195.94/kW/month nationally at year-end 2025, up 6.5% year-over-year and a new record, reflecting an 11% CAGR since 2020.(6)
Large-format deal pricing — transactions of 10 MW or more — has risen most sharply, with Silicon Valley up 19%, Chicago up 15.4%, and Northern Virginia up 13.8% in 2025.(6)
Lease structures span wholesale colocation (250 kW to 20+ MW, the most common institutional investment product), retail colocation (single racks to partial cages at higher per-kW rates), build-to-suit (custom-designed facilities, now the top opportunity identified by 49% of investors — up from 31% in 2024), and powered shell (landlord provides building and power infrastructure, tenant completes fit-out). Wholesale leases typically run 10–15 years with 2–3% annual escalations; hyperscale build-to-suit leases extend to 15–20 years.
Stabilized institutional-quality data center properties trade at 4.25–6.25% cap rates with five-year unleveraged IRRs of 7.0–8.5%.(11) CMBS-financed data centers stabilized around 6.5% in 2025, with CMBS issuance reaching an all-time high of approximately $4.5 billion in Q1 2025 alone — including a $2.4 billion Switch deal and a $2.05 billion QTS transaction.(12) Digital Realty reports stabilized yields on construction of 12.3% in North and South America, and Equinix achieves 26% cash-on-cash return on gross property, plant, and equipment — metrics that explain why 95% of investors surveyed by CBRE planned to increase data center allocations in 2025.
6. The AI Demand Supercycle
The AI infrastructure buildout is generating a capital deployment cycle with no precedent in commercial real estate history. Combined Big Four hyperscaler capital expenditures reached approximately $200 billion in 2024, with 2025 projections placing AWS at $118 billion, Google at $91–$93 billion, Meta at $70 billion, and Microsoft at approximately $80 billion.(13,14) Combined Big Five expenditures — including Oracle — are forecast at $602 billion in 2026, a 36% year-over-year increase. Goldman Sachs projects $1.15 trillion in cumulative hyperscaler capital expenditures from 2025 through 2027, of which approximately 75% is directed toward AI infrastructure.
AI represented approximately 15% of data center workloads in 2024, rising to approximately 25% in 2025, and is projected to reach 40–50% by 2030.(5) By 2027, inference workloads — the continuous computational demand of deployed AI models responding to queries — are expected to overtake training as the dominant compute requirement, sustaining demand for data center capacity well beyond the initial model development phase. The Stargate project represents $400+ billion in planned investment across nearly 7 GW, and Meta/Blue Owl's Hyperion campus alone targets 5 GW — single-project scale that exceeds the total historical installed capacity of many regional power utilities.
| Company | 2025 Projected Capex | Source |
|---|---|---|
| Amazon Web Services (AWS) | ~$118 billion | Company Earnings Guidance |
| $91–$93 billion (3rd upward revision) | Company Earnings Guidance | |
| Meta | $70 billion (2025); $115–$135B (2026) | Company Earnings Guidance |
| Microsoft | ~$80 billion | Company Earnings Guidance |
| Combined Big 4 (2024 actual) | ~$200 billion | IoT Analytics |
| Combined Big 5 (2026 forecast) | $602 billion (+36% YoY) | CreditSights, January 2026 |
| Cumulative 2025–2027 | $1.15 trillion | Goldman Sachs Research |
7. Cooling Technology and Sustainability Imperatives
The global industry average power usage effectiveness (PUE) remains at 1.56 per Uptime Institute's 2024 survey — flat for five consecutive years — though leading operators perform substantially better: Google at 1.09 and AWS at 1.15 on a fleet-wide basis.(8) The EU mandates PUE below 1.3 by 2030, and Singapore requires below 1.2 for new facilities — regulatory thresholds that will increasingly govern facility design requirements for operators with global infrastructure ambitions.
Liquid cooling has emerged as the enabling technology for AI-density infrastructure. The liquid cooling segment held 46% of the total data center cooling market in 2024 and is growing at 31.4% CAGR. Air cooling reaches its thermal ceiling at 20–40 kW per rack, while liquid cooling handles 100–200 kW per rack — a performance gap that makes liquid cooling a technical necessity rather than an option for AI-optimized facilities. Direct-to-chip cooling is projected to capture 47% of AI data center cooling by 2025, and over 50% of new hyperscale capacity is expected to be liquid-cooled by 2027.
Water consumption is an emerging regulatory and reputational risk. U.S. data centers consumed 17 billion gallons directly in 2023, a figure projected to double or quadruple by 2028.(15) Google consumed 8.1 billion gallons in 2024, nearly doubling in three years. Microsoft committed to zero water use for cooling in new designs starting August 2024, and leading operators now compete on water usage effectiveness (WUE) metrics as aggressively as on PUE — with AWS at 0.15–0.19 L/kWh, Meta at 0.20 L/kWh, and Microsoft at 0.30 L/kWh representing the current performance frontier.
8. Site Selection Criteria and Feasibility Evaluation Framework
Data center site selection is governed by a ranked hierarchy of criteria that differs fundamentally from conventional industrial or commercial feasibility. Per JLL and CBRE's 2025–2026 frameworks, the sequence is: (1) speed to power and overall power availability, (2) community support and regulatory environment, (3) latency and proximity to end users, (4) fiber connectivity and interconnection density, (5) construction cost variations by market, (6) tax incentives and economic development support, (7) available land at 200+ acres, and (8) environmental considerations including water and carbon commitments.(5,6)
Fiber connectivity remains a significant differentiator. AT&T has signed a multi-year $1 billion agreement with Corning for fiber deployment, and Zayo is constructing major fiber routes from Chicago to Columbus (385 miles) and Chicago to Minneapolis (520 miles) — infrastructure investments that are creating new viability thresholds for data center development in secondary markets along these corridors.
For feasibility studies, the evaluation framework should incorporate five core analytical components:
- power availability assessment (utility capacity, interconnection queue position, and energization timeline);
- market demand analysis (vacancy, net absorption, and active tenant pipeline);
- financial modeling (development cost per MW, projected lease rates, and stabilized yield targets of 10%+ for new development);
- site due diligence (zoning entitlement timeline, environmental, flood, and seismic assessment);
- infrastructure assessment (fiber route proximity and interconnection density).
A 30 MW facility at $10 million per MW of capital expenditure requires approximately $100 million in annual revenue to achieve a 10% unlevered internal rate of return — a benchmark that validates the economic rationale for the lease rates currently being achieved in primary and secondary markets alike. For smaller-scale colocation facilities financed through SBA or conventional debt, this framework applies at reduced scale, with particular attention to local market demand validation and anchor tenant pre-commitment as conditions precedent to lending.
9. Outlook: Power-Constrained Supply, Unconstrained Demand
The data center sector presents a feasibility paradox: demand is effectively unlimited and well-documented by the capital commitments of the world's largest technology companies, while supply creation is constrained by a single input — power — that is subject to multi-year procurement timelines, regulatory uncertainty, and grid infrastructure limitations that no amount of capital can resolve quickly. The result is a market where conventional feasibility risk assessment is inverted: absorption risk is negligible, and execution risk is concentrated entirely in the power acquisition and infrastructure development pathway.
For lenders evaluating data center project finance, the critical diligence questions are:
- What is the utility's confirmed capacity commitment and energization timeline?
- What is the project's position in the relevant interconnection queue?
- What is the pre-commitment level from anchor tenants prior to construction commencement?
- And what is the operator's track record of delivering facilities within projected cost and schedule parameters in a construction environment characterized by 36–48 week equipment lead times?
Projects that satisfy these criteria represent the highest-conviction deployment environments in commercial real estate today — and, per the investment return metrics reported by leading operators, among the most rewarding. Investment return modeling for data centers uses per-MW IRR calculations rather than the per-square-foot metrics standard in other CRE segments.
For lenders and investors evaluating projects in this sector, UnionMetric provides independent data center feasibility study services that meet SBA and USDA requirements.
For the cross-asset framework, demand regimes, valuation methods and reviewer tests, see the 2026 Feasibility Methodology Atlas.
Sources
- Precedence Research, U.S. Data Center Market Size and Forecast, 2024.
- Arizton Advisory & Intelligence, U.S. Data Center Investment Report, 2024.
- Avison Young, U.S. Data Center Market Report, Q4 2025.
- Mordor Intelligence, U.S. Data Center Capacity and Growth Forecast, 2025.
- JLL, Global Data Center Outlook 2026.
- CBRE, U.S. Data Center Trends, H2 2025.
- Cushman & Wakefield, Data Center Development Cost Guide, 2025.
- Uptime Institute, Global Data Center Survey, 2024.
- International Energy Agency (IEA), Electricity 2024 Report.
- Gartner, Data Center Power and AI Infrastructure Forecast, 2025.
- Northmarq / RCLCO, Data Center Investment and Cap Rate Analysis, 2025.
- CRE Daily / Trepp, CMBS Data Center Report, 2025.
- CreditSights, Hyperscaler Capital Expenditure Forecast, January 2026.
- IoT Analytics, Cloud and Hyperscaler Infrastructure Spending, 2024.
- Google, Environmental Report, 2025.
Looking for a lender-grade industrial feasibility study?
Related feasibility study engagements
Data Center Colocation Facility, Prince William County, Virginia
Conventional feasibility study supporting a data center project in Prince William County.
Wholesale Colocation Data Center, Ground-Up Development in a Primary Market
Conventional feasibility study supporting a data center project in Primary Data Center Market.
Michael Torres, MAI
Certified commercial real estate appraiser with extensive experience in industrial property valuations across the United States.
View team credentials →