Data Center REITs: Digital Realty and Equinix Investment Analysis

The Physical Layer of AI: Why Real Estate Matters

Every conversation about AI investing eventually turns to chips, models, or software. Almost nobody talks about the buildings. This is a mistake. The physical data center — the concrete, steel, copper, and cooling infrastructure that houses the GPUs — is the scarcest and most capital-intensive component of the AI stack. An NVIDIA H100 chip costs $25,000 and can be shipped overnight. The data center that houses it costs $8–12 million per megawatt to build and takes 18–36 months to construct.

The numbers are extraordinary. CBRE estimates that global data center construction starts reached $80B in 2024 and will exceed $100B annually by 2026. JLL reports that US data center absorption set records in 9 of the last 12 quarters. Primary markets — Northern Virginia (the world's largest data center cluster), Dallas, Phoenix, Chicago, and Silicon Valley — have vacancy rates below 3%. In Northern Virginia, functional vacancy is essentially zero; every megawatt of new capacity is pre-leased before construction begins.

For investors, this translates to a simple framework: the companies that own and control data center capacity are in a landlord's market with no foreseeable end. When supply takes 2–3 years to build and demand is growing at 25–30% annually, the pricing power accrues to the asset owners — the REITs.

This is not a new observation, but the market has been slow to fully price it. Data center REITs are classified in the Real Estate sector by GICS, which means they are evaluated by real estate analysts, held in real estate ETFs, and benchmarked against office buildings and shopping malls. The result is a persistent categorization discount: these are technology infrastructure assets trapped in a real estate valuation framework.

Equinix (EQIX): The Interconnection Premium

Equinix is the world's largest data center REIT with a market capitalization of approximately $85B. It operates over 260 data centers across 72 metropolitan areas in 33 countries. But characterizing Equinix as merely a data center landlord fundamentally misunderstands the business. Equinix is an interconnection company that happens to operate data centers.

The Interconnection Moat

Equinix's competitive advantage is network density. Its data centers are the physical locations where enterprises, cloud providers, content delivery networks, and telecommunications carriers interconnect — literally plugging cables between their equipment to exchange data. This creates a powerful flywheel: the more networks present in an Equinix facility, the more valuable it becomes for every additional customer. A new enterprise customer does not choose Equinix because the building is nicer than a competitor's. They choose Equinix because 2,000 other companies are already there, and connecting to them inside the facility is faster, cheaper, and more reliable than doing so over the public internet.

This network effect produces extraordinary economics. Equinix's interconnection revenue — the fees charged for cross-connects, fabric ports, and internet exchanges — carries gross margins estimated above 90% and has grown at 10–12% annually for over a decade. Interconnection represents approximately 20% of total revenue but likely over 35% of gross profit. The revenue is deeply recurring (average cross-connect tenure exceeds 7 years), high-margin, and virtually impossible for competitors to replicate because moving interconnection requires every connected party to coordinate simultaneously.

Equinix Financial Profile

For FY2025, Equinix guided for revenue of $8.8–8.9B (8–10% growth), AFFO per share of $36–37 (7–9% growth), and a 3% dividend increase to approximately $17.50 per share. The company has delivered 22 consecutive years of revenue growth — through the GFC, COVID, and every rate cycle in between. AFFO per share has compounded at 10%+ annually for over a decade.

At the current price of approximately $860, Equinix trades at roughly 24x forward AFFO. This is above its 10-year average of 22x but below the 28x peak reached in 2021. The premium is justified by the AI-driven acceleration in leasing activity: Equinix reported record new leasing bookings in Q3 and Q4 2024, with particularly strong demand for its xScale hyperscale joint venture facilities.

Digital Realty (DLR): The Hyperscale AI Landlord

Digital Realty is the second-largest data center REIT at approximately $55B market cap, operating over 300 data centers across 50+ metropolitan areas in 25+ countries. While Equinix dominates interconnection, Digital Realty is the hyperscale landlord — the company that builds the massive, purpose-built facilities that house the GPU clusters training the world's largest AI models.

The Hyperscale Leasing Machine

Digital Realty's hyperscale leasing has entered a phase that management describes as “unprecedented.” In 2024, the company signed a record $1.4B in new leases, including multiple deals above 40 MW — the size that AI training facilities require. These leases are typically 10–15 years in duration with 2–3% annual escalators and power pass-through provisions that protect DLR from energy cost volatility.

The pricing dynamic is the most important story. In 2022, hyperscale lease rates in Northern Virginia — the bellwether market — were approximately $120–140 per kW per month. By late 2024, rates for AI-ready capacity (higher power density, liquid cooling infrastructure, redundant utility feeds) had risen to $170–200+ per kW per month. That is a 40–60% increase in 24 months for what is essentially the same square footage. The difference is power density: an AI-ready cabinet consuming 40 kW generates 4–5x the revenue per square foot as a standard 8 kW cabinet.

Digital Realty's development pipeline — the projects under construction that will generate future revenue — exceeded 700 MW at the end of 2024, the largest backlog in company history. At stabilized yields of 8–11% on development cost, this pipeline represents $4–6B in annual recurring revenue once delivered and stabilized over the next 2–4 years.

Digital Realty Financial Profile

Digital Realty guided for 2025 revenue of $6.0–6.1B (6–8% growth), core FFO per share of $7.00–7.10 (5–7% growth), and maintained its $4.88 annual dividend ($1.22/quarter). The dividend yield of approximately 2.9% provides income while investors wait for the development pipeline to translate into AFFO growth.

At approximately $170 per share, Digital Realty trades at roughly 24x forward FFO. The valuation re-rating from the 2022 trough of 15x reflected the market's recognition of AI-driven demand, but we believe the lease rate expansion and development pipeline are not yet fully reflected in consensus estimates. If hyperscale rates sustain at $180+ per kW and the development pipeline delivers at 10% yields, Digital Realty's AFFO could grow at 10–12% annually through 2028 — well above the 5–7% that current guidance implies.

The Supply-Demand Equation: Why This Cycle Is Different

Skeptics will point to past data center construction cycles that ended in oversupply and compressed returns. The 2000–2002 fiber-optic bubble and the 2015–2017 commodity colocation oversupply in Northern Virginia are cautionary precedents. So why do we believe this cycle is structurally different?

Three constraints prevent the traditional supply response. First, power availability. In Northern Virginia, Dominion Energy has stated that new utility substations require 3–5 years of lead time. In Dublin, Ireland (Europe's largest data center market), the government has imposed a moratorium on new data center connections to the grid until 2028. In Singapore, new data center construction was frozen for 3 years before a limited reopening in 2024. Power is not a matter of capital — you can raise money to build a data center in months, but you cannot conjure a 100 MW utility connection.

Second, land constraints. Data centers require specific site characteristics: proximity to fiber routes and internet exchanges, access to redundant utility feeds, adequate water supply for cooling, and favorable natural disaster risk profiles. In tier-one markets, the supply of suitable land is finite and increasingly contested. Equinix and Digital Realty have spent decades assembling land banks that newer competitors cannot easily replicate.

Third, the AI power density shift. Traditional cloud workloads consume 5–10 kW per rack. AI training consumes 30–50 kW per rack, and next-generation clusters with liquid cooling may reach 100+ kW. Existing data centers built for traditional density cannot simply “upgrade” to AI density — the electrical infrastructure, cooling systems, and structural engineering are fundamentally different. This means that a significant portion of the existing 30+ GW of global data center capacity is functionally obsolete for AI workloads, tightening the effective supply of AI-ready capacity even further.

For more on the broader AI infrastructure supply chain, our analysis of nuclear energy stocks powering AI data centers covers the power generation side of this equation, while our piece on AI infrastructure investment trends provides the full capex framework.

Risks: The Bear Case for Data Center REITs

We are constructive on data center REITs but intellectually honest about the risks. The most frequently cited concern — interest rate sensitivity — is actually the least interesting risk. Yes, REITs are capital-intensive and rely on debt financing. Yes, higher rates increase the cost of capital. But data center REITs fund development at 8–12% unlevered yields with debt at 4–5%. The spread is healthy and improving as lease rates rise.

The more material risks are structural. Customer concentration is a genuine concern. The top 20 hyperscale tenants represent 30–40% of Digital Realty's revenue, and these customers have enormous bargaining power. If Microsoft or Google decided to build more capacity in-house rather than leasing from REITs, the growth trajectory would decelerate. Meta's $40B 2025 capex plan includes significant owned data center construction. The question is whether hyperscalers will continue to supplement their owned capacity with leased capacity from REITs, or whether the leased model becomes secondary over time.

AI efficiency improvements represent a longer-term structural risk. DeepSeek's V4 model demonstrated that competitive AI performance can be achieved with significantly less compute. If AI inference becomes dramatically more efficient — requiring fewer GPUs and therefore less data center capacity per unit of AI output — the demand projections that underpin data center REIT valuations could prove overly aggressive. We believe this is a real risk for the 2028–2030 timeframe, though we expect efficiency gains to be offset by demand expansion (Jevons paradox) in the near term.

Environmental regulation is an emerging risk that the market underappreciates. Data centers consume massive amounts of water for cooling — a single large facility can use 1–5 million gallons per day. In water-stressed regions (Phoenix, Las Vegas, parts of Texas), local governments are beginning to push back on data center water permits. If regulations force a shift to air-cooled or immersion-cooled designs, retrofit costs could be significant for existing facilities.

Portfolio Positioning: How to Own the Physical Layer

For investors seeking AI infrastructure exposure with the stability of contracted cash flows, we recommend a core allocation to data center REITs of 3–6% of a diversified equity portfolio. The allocation should be viewed as a 3–5 year holding, not a tactical trade, because the thesis depends on the conversion of record lease bookings into recognized revenue over time.

We believe the optimal approach is to own both Equinix and Digital Realty, weighted approximately 55/45 in favor of Equinix for portfolios prioritizing defensive quality and 45/55 in favor of Digital Realty for portfolios prioritizing direct AI demand leverage. The two companies are more complementary than competitive — Equinix dominates interconnection, Digital Realty dominates hyperscale — and holding both diversifies across the different segments of data center demand.

Beyond the two majors, investors seeking additional exposure should consider CyrusOne (privately held by KKR, a potential re-IPO candidate), CoreWeave (recently IPO'd, pure-play AI cloud infrastructure), and Switch (SWCH, acquired by DigitalBridge in 2022, potential return to public markets). The data center REIT ecosystem is consolidating, and M&A activity has been significant — another potential catalyst for valuation re-rating of the remaining public players.

What to Monitor Going Forward

The key data points to track for data center REITs are: quarterly new lease signings (volume and pricing), same-store occupancy trends (should remain above 90%), hyperscale renewal spreads (the pricing increase on lease renewals), development pipeline size and yield, power availability updates in primary markets, and hyperscaler capex guidance (each earnings season from MSFT, GOOGL, META, and AMZN). These data points are often buried deep in earnings call transcripts and supplemental financial packages — exactly the kind of unstructured information that AI-powered analysis tools excel at extracting systematically.

Frequently Asked Questions