How Are AI Infrastructure Demands Reshaping Data Center REIT Fundamentals?

Data center REITs own the physical facilities housing computing infrastructure — the buildings, power systems, cooling systems, and network connectivity that enable cloud computing, enterprise IT, and AI training workloads. The sector has been fundamentally transformed by the AI infrastructure buildout: hyperscalers (Microsoft, Amazon, Google, Meta) have accelerated data center construction spending to hundreds of billions annually, creating unprecedented demand for colocation facilities and power infrastructure. Equinix and Digital Realty — the two dominant public data center REITs — occupy different market positions (retail colocation versus hyperscale wholesale) that expose them differently to the AI demand surge and to the risk of hyperscaler direct ownership.

Quick definition: Data center facility types: (1) Retail colocation — customers lease individual cages or cabinets (1–50 kilowatt range) in Equinix's carrier-neutral facilities; interconnection between thousands of networks is the primary value driver; (2) Hyperscale wholesale — customers (hyperscalers) lease entire buildings or large floors (1–100+ megawatts) under long-term contracts; Digital Realty's primary product; (3) Edge computing facilities — smaller facilities close to end users for low-latency applications; (4) Powered shells — buildings with power, cooling, and shell; tenant installs equipment and finishing; (5) Build-to-suit — custom facilities built to tenant specification, then sold or leased back.

Key takeaways

• Equinix's interconnection ecosystem (the network of 10,000+ customers exchanging traffic within its facilities) creates a genuine switching cost moat — when a company has direct connections to hundreds of partners, customers, and cloud providers within an Equinix facility, moving to a competing facility would require rebuilding all those physical connections; this interconnection density is the primary reason Equinix commands 10–15% premium rents versus comparable competing facilities
• Power availability has replaced capital as the primary constraint on data center supply growth — US electrical grid capacity limitations (transformer lead times of 24–48 months, transmission permitting of 5–10 years) mean that data center developers cannot simply build where land is cheapest; markets with available power (Northern Virginia, Dallas, Phoenix, Atlanta) attract development disproportionately to available electrical infrastructure
• Hyperscalers building their own proprietary data centers (rather than leasing from REITs) create competition for Digital Realty's wholesale business but not for Equinix's colocation business — Amazon's AWS, Microsoft Azure, and Google Cloud all own substantial proprietary data center capacity; however, they also purchase capacity from Equinix and Digital Realty for networking peering, geographic coverage, and temporary capacity; this hybrid model means hyperscaler capex growth is partially complementary to and partially competitive with data center REITs
• Digital Realty's partnership with Blackstone for its data center developments (DataBank, QTS) illustrates the capital intensity challenge — building hyperscale data centers (50–200+ MW each) requires billions in investment; forming joint ventures and platforms with private capital partners enables scale without individual REIT balance sheet concentration
• AI GPU cluster infrastructure has different power density requirements than traditional servers — AI training clusters can consume 40–100+ kilowatts per rack versus 8–15 kW per rack for traditional compute; this higher power density means the same building footprint accommodates fewer racks but more revenue per square foot, benefiting operators who can provide high-density power infrastructure

Equinix business model analysis

Interconnection revenue: Equinix operates 260+ International Business Exchange (IBX) data centers across 70+ cities in 32 countries. Its business model has two revenue streams: (1) colocation (space, power, cooling) — approximately 75–80% of revenue; (2) interconnection — fees for direct connections between customers within Equinix facilities — approximately 20–25% of revenue and growing faster than colocation. Interconnection revenue has high margins (incremental port revenue with minimal incremental cost) and high renewal rates (customers who have built connection graphs across hundreds of partners do not leave easily).

Recurring revenue quality: Equinix's revenue is approximately 95% recurring (monthly recurring revenue from colocation and interconnection contracts) with average contract terms of 2–3 years. The combination of recurring revenue, interconnection switching costs, and geographic diversification (North America 45%, EMEA 35%, Asia-Pacific 20%) produces earnings quality comparable to regulated utilities — predictable, contractual, and resistant to sudden demand shocks.

Network density moat: Equinix's facilities contain 10,000+ customers (enterprises, cloud providers, financial services, content companies, carriers). When a company needs to exchange traffic with partners (interconnect), the probability of those partners being in an Equinix facility is very high — creating a network effects dynamic. Adding a new customer to an Equinix IBX makes the facility more valuable to all existing customers (more potential connection partners), reinforcing the density advantage over time.

How it flows

Digital Realty wholesale model

Hyperscale customer concentration: Digital Realty's top customers include Microsoft, Meta, IBM, Oracle, and major cloud/telecom providers — large enterprise and hyperscaler customers leasing significant capacity (multiple megawatts to multiple buildings) on 5–15 year contracts. This long-term lease structure provides revenue predictability but creates lease expiration concentration risk — when a hyperscale customer's lease expires (or if a major customer reduces capacity), Digital Realty faces significant re-leasing challenge.

Churn risk differentiation: Digital Realty's wholesale business has higher churn risk than Equinix's retail colocation — hyperscalers who own their own data center infrastructure can reduce REIT exposure as they build proprietary capacity. In practice, hyperscalers both build proprietary facilities AND purchase colocation/wholesale capacity from REITs — for geographic coverage, temporary capacity, and network interconnection. The hyperscaler relationship is therefore partially competitive and partially complementary.

PlatformDIGITAL framework: Digital Realty markets a "PlatformDIGITAL" strategy — providing standardized global data center infrastructure that enables customers to deploy consistently across Digital Realty facilities worldwide. This standardization reduces customer IT complexity for global deployments and creates stickiness by making Digital Realty facilities functionally interchangeable — customers who build their deployment architecture on PlatformDIGITAL standards prefer Digital Realty globally over assembling a patchwork of local operators.

Power constraint as competitive moat

Grid capacity as development bottleneck: The single most important constraint on data center development in 2024–2028 is available electrical power. Dominion Energy's Northern Virginia grid (the dominant US data center market) has signed power commitments for load growing from approximately 3.5 GW to potentially 8–12 GW by 2030 — and the limiting factor is transmission capacity and transformer delivery, not developer capital or demand. Markets where grid expansion is underway (Northern Virginia, Dallas, Phoenix) will see supply growth; markets where grid capacity is constrained (some Bay Area markets) will see limited supply regardless of demand.

Data center REITs as power infrastructure: Given power constraint, data center REITs with secured power (grid interconnection agreements, dedicated substation capacity, backup generator fleets) have a competitive advantage over new entrants attempting to secure power in constrained markets. Equinix and Digital Realty have secured power allocations in their existing markets through decades of relationships with utilities — making their expansion of existing campus footprints easier than greenfield development in new locations.

Iron Mountain's data center transition

Iron Mountain (IRM): Historically the dominant physical records management company (storing paper documents and media for compliance purposes), Iron Mountain has been building a data center business by converting its urban industrial facilities (many in city centers ideal for low-latency data center use) into colocation capacity. Iron Mountain's data center segment has been growing rapidly — representing approximately 20–30% of EBITDA and growing toward 50% as the traditional storage business slowly declines.

Strategic repositioning: Iron Mountain exemplifies a REIT repositioning toward growth by pivoting its asset base from declining (physical paper storage) toward growing (data center colocation) property types. The conversion of urban storage facilities with existing power connections and structural reinforcement into data centers is more cost-effective than greenfield development in some markets. For investors, evaluating Iron Mountain requires separating the declining legacy business from the growing data center business — a sum-of-parts analysis that aggregate REIT metrics obscure.

Common mistakes

Conflating Equinix's colocation model with Digital Realty's wholesale model. These are fundamentally different businesses: Equinix's interconnection moat and retail colocation provide different risk/return characteristics from Digital Realty's hyperscale wholesale direct competition exposure. Owning both provides diversification within data centers; treating them as interchangeable substitutes misses the business model differentiation.

Assuming AI demand growth directly translates to data center REIT revenue growth. AI training clusters have specific power density, cooling (often liquid cooling rather than air cooling), and location requirements that not all existing data center infrastructure can serve. Equinix's traditional retail colocation facilities may require significant upgrade to serve GPU-dense AI workloads; Digital Realty's newer hyperscale builds can more readily accommodate AI density. AI demand is a tailwind, but not uniformly applicable to all data center facility types.

FAQ

How does the competition between hyperscaler owned data centers and data center REITs evolve over time?

The hyperscaler-owned versus REIT-leased data center tension has been debated for 15 years — and the outcome has been coexistence rather than displacement. Amazon AWS, Microsoft Azure, and Google Cloud own massive proprietary data center capacity while simultaneously being among the largest customers of Equinix and Digital Realty. The reasons for coexistence: (1) interconnection requirements — neutral facilities where all networks exchange traffic cannot be replaced by proprietary hyperscaler campuses; (2) geographic coverage — 260+ Equinix locations globally would require hundreds of billions to self-build; (3) enterprise customer needs — enterprise IT buyers need colocation where their private infrastructure meets hyperscaler connections; (4) capacity overflow — hyperscaler construction pipelines cannot keep pace with AI-driven demand growth, making REIT-leased capacity necessary for near-term needs. The trend is more hyperscaler self-build (reducing the wholesale data center REIT addressable market percentage) while absolute volume grows with overall demand. CBRE's data center market data at cbre.com; Datacenter Dynamics tracks global data center construction activity at datacenterdynamics.com.

Related concepts

• Real Estate Overview
• Industrial REITs
• REIT Valuation
• Cell Tower REITs
• Real Estate Portfolio Sizing

Summary

Data center REITs occupy two distinct market positions: Equinix's colocation/interconnection business (strong network effects moat, 10,000+ customer density, interconnection switching costs) and Digital Realty's wholesale hyperscale business (long-term contracts, hyperscaler customer concentration, direct competition risk from hyperscaler owned capacity). Power availability — not capital — is the primary data center supply constraint, giving existing operators with secured grid capacity a durable competitive advantage over new entrants in constrained markets. AI GPU cluster infrastructure (40–100 kW/rack power density) creates a specific demand signal for high-density power-capable facilities — a retrofit and upgrade challenge for older facilities but a premium opportunity for operators who can deliver. Iron Mountain's strategic pivot from physical records storage to data center colocation illustrates the REIT asset base repositioning possible through facility conversion. Hyperscaler self-build competition is real but coexists with REIT leasing — interconnection requirements, geographic coverage economics, and overflow demand necessitate continued REIT capacity alongside hyperscaler proprietary investment.

Next

→ Residential REITs: Apartments, Single-Family Rentals, and Housing Affordability