Key Takeaways
- Institutional capital has decisively rotated into data centers: 95% of major investors plan to raise allocations, and Blackstone alone holds $70 billion in data center assets—making it the world's largest provider.
- Power, not land, is the new supply constraint. Grid connection wait times in primary markets exceed four years, making electrical infrastructure the single most decisive factor in site selection and asset valuation.
- Northern Virginia absorbed 1,102 MW in 2025—more than most legacy gateway cities transact in total CRE volume—while Dallas absorbed 471 MW, cementing a winner-take-all geographic concentration.
- Construction costs rose at a 7% CAGR from 2020–2025 (reaching $10.7M/MW), and JLL projects another 6% increase in 2026, embedding durable pricing power into stabilized assets.
- Data centers delivered 11.2% returns in the most recent measurement period, outperforming office, retail, and most multifamily, while cap rates compressed from ~6.5% in 2023 to 5.0–5.5% for hyperscaler-occupied assets by 2026.
The U.S. mall market peaked at roughly 1,500 enclosed malls. Today, fewer than 700 are viable. Meanwhile, the global data center sector is on track to add nearly 100 GW of new capacity between 2026 and 2030—doubling global capacity and creating an estimated $1.2 trillion in real estate asset value in the process. This is not a niche asset class disrupting at the margins. It is a structural realignment of where institutional capital believes long-duration, income-generating real estate value actually lives.
The thesis is simple and brutal: AI has permanently elevated data centers from a cyclical niche into the definitive core real estate asset of this investment cycle—and the institutions that fail to recategorize their portfolio frameworks accordingly will misprice both opportunity and risk for the rest of the decade.
Why Morgan Stanley, Apollo, and JPMorgan Are All Saying the Same Thing About Data Centers
When three of the largest capital allocators on the planet converge on a single thesis, the signal is worth heeding. JPMorgan's analysis pegs global data center and AI infrastructure spend at $5 trillion through 2030, with 122 GW of new capacity expected between 2026 and 2030. Their conclusion: "demand for compute remains astronomical," and annual financing needs of approximately $700 billion can be absorbed entirely through hyperscaler cash flows and investment-grade debt markets. Morgan Stanley's 2026 Infrastructure Outlook identifies data centers and power infrastructure as the two most consequential trends reshaping portfolio construction. Apollo's framework mirrors this, flagging digital infrastructure as the primary vehicle for risk-adjusted return capture as traditional CRE sectors face structural headwinds.
This is not enthusiasm for a trend—it is a coordinated repricing of what "core" real estate means. The Magnificent Seven tech companies alone are expected to deploy $414 billion in AI and data center capital expenditures in fiscal 2025, rising to over $430 billion in 2026. That demand pipeline is longer, more contracted, and more credit-worthy than anything the office or retail sectors have produced in the last 15 years.
The Three Geographic Clusters That Will Capture Most of the Institutional Capital
Data center investment is hyperconcentrated by design—proximity to fiber routes, power infrastructure, and talent pools creates entrenched advantages that compound over time. Three markets are pulling disproportionately far ahead.
Northern Virginia absorbed 1,102 MW in 2025, making it the single largest data center market on the planet by a substantial margin. Dallas absorbed 471 MW—up 424 MW year-over-year—as hyperscalers rotate capacity toward markets with more permissive power dynamics. Chicago maintains vacancy below 3.1%, reflecting acute demand pressure and a shortage of large contiguous blocks of space.
Think of these three clusters the way institutional capital once thought about gateway cities for office and multifamily: they are the tier-one allocation targets, the markets where liquidity is deepest and tenant covenants are strongest. Secondary markets—Atlanta, Dallas-Fort Worth's outer ring, West Texas—are attracting development capital, but stabilized institutional-grade product remains concentrated in the primary three. The geographic concentration is not accidental; it is the product of decades of fiber investment, utility relationships, and hyperscaler operational familiarity that cannot be replicated quickly in greenfield locations.
Power Constraints: The Hidden Choke Point That Makes Location Everything
The conventional real estate supply constraint is land. In data centers, the binding constraint is electrons. According to Green Street analyst David Guarino, "the main limit in today's data center market is not tenant appetite but access to power." That statement redefines how to underwrite location value in this asset class.
In Northern Virginia, Dominion Energy has publicly acknowledged that grid capacity cannot match demand. Large-load interconnection timelines in constrained primary markets can stretch to seven years. JLL reports that average grid connection wait times in primary markets already exceed four years. The result is a structural supply ceiling that cannot be built through quickly—reinforcing the pricing power of existing stabilized assets and rewarding operators who locked up power agreements years in advance.
This dynamic means data center underwriting now requires a power audit as rigorous as a title search. Sites with behind-the-meter generation, dedicated substations, or firm power purchase agreements trade at meaningful premiums. Markets where utility queues are shorter—Dallas, Phoenix's exurbs, select Midwest locations—are drawing development capital precisely because the power path to delivery is shorter, not because the real estate fundamentals are superior. Power access is the new zoning.
How Data Centers Compare to Office, Retail, and Multifamily on a Risk-Adjusted Basis
Data centers delivered 11.2% returns in the most recent measurement period, second only to manufactured housing among institutional real estate sectors. Cap rates for stabilized, hyperscaler-occupied assets compressed from approximately 6.5% in 2023 to 5.0–5.5% by 2026—a trajectory that mirrors what happened to Class-A industrial in the early 2010s as e-commerce rewrote distribution economics.
Contrast this with the broader CRE landscape. Office is still working through a multi-year value correction, having only troughed in mid-2025. Retail transaction volume recovered to approximately $60 billion in 2025—a 27% year-over-year improvement—but this rebound reflects distressed repricing and necessity-driven demand, not structural growth. Multifamily is stabilizing, but rent growth is geographically bifurcated and Sun Belt markets that overbuilt in 2021–2023 are still absorbing excess supply.
For a pension fund or endowment seeking long-duration, inflation-linked income, data centers now offer what office once promised: long lease terms (hyperscaler contracts commonly run 10–15 years), creditworthy tenants, and operating fundamentals insulated from consumer sentiment. The difference is that data center demand is not cyclical—it is exponential. CBRE's 2025 Global Data Center Investor Intentions Survey found that 95% of major institutional investors plan to raise allocations to the sector, a level of consensus that has no parallel in traditional CRE.
What a 7% Construction Cost CAGR Actually Means for Portfolio Construction
Average global data center construction costs rose from $7.7 million per MW in 2020 to $10.7 million per MW in 2025—a 7% CAGR—and JLL projects a further 6% increase in 2026, pushing costs to $11.3 million per MW. For AI-optimized facilities with custom tenant fit-outs, costs can reach $25 million per MW. Simultaneously, the sector is expected to expand at a 14% CAGR through 2030, with AI representing half of all data center workloads by the end of the decade.
Sustained cost escalation is not a margin compression story—it is a moat. Rising replacement costs systematically increase the value of existing stabilized assets, particularly those in power-constrained primary markets where new supply cannot be brought online at any cost within a reasonable timeline. This is the same dynamic that made pre-2010 Class-A industrial in infill markets so valuable once e-commerce demand exceeded replacement cost thresholds. For portfolio construction, entry timing matters enormously, and development exposure carries a fundamentally different risk profile than stabilized acquisition.
The New Definition of 'Core' Real Estate — and What Gets Left Behind
Blackstone's A$24 billion acquisition of AirTrunk—the largest data center deal in history—was structured across four separate Blackstone strategies: real estate, infrastructure, tactical opportunities, and private equity. That cross-vehicle architecture is deliberate and revealing. Blackstone now holds $70 billion in data center assets and over $100 billion in prospective pipeline, making it the world's largest data center provider. The signal to the market is unambiguous: data centers are not an alternative asset class supplement to real estate—they are the primary real estate asset class of this cycle.
The institutional real estate industry is still arguing about whether data centers belong in the real estate bucket or the infrastructure bucket. That definitional debate is costing LPs real return. Funds constrained to traditional CRE mandates are watching specialists accumulate the most durable income streams in the asset class while arguing about taxonomy.
The practical implication: "core" real estate in 2030 will look nothing like "core" real estate in 2015. Gateway office towers and regional mall anchors were the definitive core assets of the last cycle. This cycle's core assets are the hyperscale campuses in Ashburn, Virginia, and the data hall clusters outside Dallas being pre-leased by Microsoft and Google on decade-long terms before a single foundation is poured. McKinsey projects $5.2 trillion in global data center infrastructure investment by 2030. Capital that waits for definitional consensus will have missed the repricing entirely.
Frequently Asked Questions
Are data center REITs still a viable entry point given current cap rate compression?
Cap rates for stabilized, hyperscaler-occupied data centers have compressed to 5.0–5.5% by 2026, down from approximately 6.5% in 2023, making stabilized acquisitions expensive relative to historical spreads. However, 62% of institutional survey respondents favor opportunistic new development strategies over core acquisitions, per CBRE's 2025 Global Data Center Investor Intentions Survey, suggesting the better risk-adjusted entry is development exposure in markets with shorter power delivery timelines rather than stabilized asset acquisition.
How does power constraint in Northern Virginia affect asset valuations there?
Northern Virginia remains the world's largest data center market with 1,102 MW absorbed in 2025, but Dominion Energy has acknowledged grid capacity cannot match demand, and large-load interconnection timelines can reach seven years. This supply ceiling structurally supports existing asset values—assets with secured power agreements or behind-the-meter generation trade at meaningful premiums—but new development is increasingly pushed to surrounding counties and adjacent states where utility queues are shorter.
What separates hyperscale data centers from colocation facilities as investment vehicles?
Hyperscale facilities—purpose-built for a single tenant like Microsoft, Google, or Amazon—offer long-term (10–15 year) leases with investment-grade covenants, providing income certainty comparable to net-lease industrial assets but with construction costs now reaching $25 million per MW for AI-optimized facilities, per JLL's 2026 Global Outlook. Colocation assets serve multiple tenants and offer more diversification but shorter lease terms and higher management complexity; colocation is growing at 19% in APAC markets where enterprise cloud migration is still accelerating.
Will secondary data center markets like Atlanta or West Texas displace primary clusters for institutional capital?
Secondary markets are attracting development capital because power delivery timelines are shorter and permitting more permissive than in constrained primary markets, but they will not displace Northern Virginia, Dallas, or Chicago for stabilized institutional investment. The fiber density, talent concentration, and existing hyperscaler infrastructure in primary clusters create network effects that compound over time; secondary markets will capture development volume but price at meaningful cap rate discounts to primary assets for the foreseeable future.
How should institutional LPs think about the $5 trillion AI infrastructure spend figure in terms of real estate exposure?
JPMorgan's $5 trillion global data center and AI infrastructure spend projection through 2030 encompasses IT equipment, power infrastructure, and tenant fit-out—not just real estate. JLL estimates the 100 GW capacity addition creates $1.2 trillion in real estate asset value, with approximately $870 billion in debt financing required, representing the largest single-sector CRE debt opportunity in history and a significant driver of CMBS and private credit volume over the coming years.