Do Data Center REITs Give You AI Exposure? A Closer Look

REITs own the buildings AI runs in, but not the AI business itself, here's what that distinction means for returns.

Prepared by Richstorm.co

Key Takeaways

• The largest publicly traded data center REITs, Equinix, Digital Realty, and Iron Mountain, returned roughly 39 to 45 percent over the twelve months to April 2026, but this still trailed the AI companies and chipmakers whose demand drives that performance.

• The largest AI training facilities are built and owned by the hyperscalers themselves; REITs primarily capture AI-adjacent demand through build-to-suit leasing, scarce power and land access, and a growing but still modest share of AI inference workloads.

• REITs earn a landlord's share of the value chain, rental income on infrastructure, while the tenants running AI workloads capture the far larger revenue from the compute itself, a structural gap that recent returns reflect.

• Broad REIT index funds such as Vanguard's VNQ have returned about 5 percent annually over the past decade versus roughly 14 percent for the broader stock market, and dilute any AI-related exposure further by including apartments, malls, and other unrelated property types.

• REITs are also interest-rate sensitive, and the most recent inflation reading of 4.2 percent, accelerating for a third consecutive month, weakens the case for the Fed rate cuts that a REIT recovery thesis depends on.

• For an investor specifically seeking exposure to AI's profitability, direct positions in hyperscalers, chipmakers, or a technology-focused fund capture that value more directly than data center REITs; REITs are better understood as an income-oriented diversifier than as an AI investment.

What a REIT Is, and How to Access One

A real estate investment trust is a company that owns, operates, or finances income-producing real estate and is required to distribute at least 90 percent of its taxable income to shareholders as dividends. This structure lets ordinary investors own a share of large commercial real estate portfolios, office buildings, warehouses, data centers, without buying property directly, and REITs trade on stock exchanges like ordinary stocks.

There are two ways to access data center REITs. The first is buying individual REIT stocks directly, Equinix (EQIX), Digital Realty (DLR), and Iron Mountain (IRM) are the three largest publicly traded names with meaningful data center exposure, alongside American Tower (AMT), a cell tower REIT with growing data center operations. The second is through REIT funds: broad real estate ETFs such as Vanguard's Real Estate ETF (VNQ) hold a diversified basket across all property types, including these names but diluted by apartments, malls, hospitals, and other categories, while more targeted funds such as the Pacer Benchmark Data and Infrastructure Real Estate SCTR ETF (SRVR) concentrate specifically on data center and digital infrastructure REITs. Buying the individual names is effectively stock-picking within the sector; buying a broad fund trades concentration for diversification, at the cost of diluting whatever data center or AI theme motivated the purchase in the first place.

The Three Names Behind the AI Infrastructure Narrative

Equinix operates more than 270 data centers across 77 global markets, with over 507,000 interconnections across more than 10,500 customers. Its business model centers on interconnection, the physical cross-connects linking enterprises, cloud providers, and telecom carriers within its facilities, generating recurring revenue with gross margins above 90 percent. Equinix is investing $4 to 5 billion annually from 2026 through 2029, targeting a doubling of its data center capacity by 2029.

Digital Realty operates more than 300 data centers for over 5,000 customers across more than 55 global markets, with a tenant list including Microsoft Azure, AWS, Nvidia, Google Cloud, Oracle, IBM, and others. Its business is more hyperscale-focused than Equinix, oriented toward the large, wholesale capacity leases that AI training infrastructure requires.

Iron Mountain, historically a records and document storage company, has pivoted aggressively toward data centers. Its fourth-quarter 2025 revenue of $1.84 billion grew 16.6 percent year over year, the fastest among the three, with data center revenue surging 39.1 percent and its asset lifecycle management business, which handles decommissioned servers as hyperscalers upgrade hardware, growing 70 percent. Iron Mountain has announced 400 megawatts of new capacity coming online over the next two years specifically targeting AI inference workloads, and raised its dividend for the fourth consecutive year.

Over the twelve months to April 2026, these three returned roughly 39, 41, and 45 percent respectively, outperforming the broader REIT sector. The obvious question is whether this represents genuine AI infrastructure ownership, or exposure to AI-adjacent demand one step removed from where the value is actually created.

Who Actually Builds AI Data Centers

The large-scale AI training facilities, the ones running tens of thousands of Nvidia GPUs at 30 to 100-plus kilowatts per rack with full liquid cooling, are predominantly built and owned by the hyperscalers themselves. The Stargate campus in Abilene, Texas, a $100 billion OpenAI, Oracle, and SoftBank collaboration targeting 5 gigawatts of capacity and more than 2 million AI chips, is owned by its builders, not leased from a REIT. Hyperscalers both design, build, and operate their own data centers and contract with colocation providers for additional capacity, meaning REIT involvement exists, but is not the primary vehicle for AI training infrastructure.

Where REITs do participate is through build-to-suit arrangements, designing, building, and operating a facility for a single tenant, often a hyperscaler needing capacity faster than its own construction pipeline allows, and through AI inference workloads, which are less power-intensive than training and more compatible with upgraded colocation space. Iron Mountain's 400-megawatt inference-focused buildout is the clearest example. Existing REITs also benefit from a scarcity premium unrelated to their buildings: REITs with grid access and power purchase agreements secured before the AI boom command premium valuations because they control scarce electrical capacity, not because their facilities are AI-optimized.

The practical upshot is a three-way split. AI training, the most capital-intensive and highest-value workload, sits almost entirely with hyperscalers building their own infrastructure. AI inference is a growing but still modest share of REIT business. Everything else, conventional cloud, enterprise IT, and other non-AI workloads, remains the bulk of REIT revenue and is unaffected by the AI buildout one way or the other.

Landlord Economics: Why the Gap Exists

When a hyperscaler builds and operates its own AI data center, it captures the entire value chain: the compute revenue from customers running AI workloads, the margin on AI services, and the appreciation of the underlying infrastructure. A REIT leasing land or a building to that hyperscaler captures only the rental income, a relatively thin slice of the value created inside that building. Construction costs for AI-optimized facilities run $20 million or more per megawatt due to liquid cooling and high-voltage power distribution; the rental income a REIT earns on that investment is a small fraction of the revenue a hyperscaler generates running AI workloads in the same space.

This is the structural reason data center REIT returns, strong as they have been, still trail the companies actually running AI workloads and the chipmaker whose products make those workloads possible. A REIT is a lower-risk, income-generating way to have some exposure to the infrastructure buildout, predictable rental income under long-term contracts regardless of whether AI itself proves profitable for its operators, but it is structurally not positioned to capture AI's upside the way the operators and chip suppliers are.

The Broad REIT Picture: VNQ Versus the Market

Stepping back from data center REITs specifically to the broad REIT category illustrates the dilution problem clearly. Vanguard's Real Estate ETF (VNQ), the largest REIT fund, has delivered annual total returns of about 5.1 percent over the past decade, compared with roughly 14.3 percent for the broader US stock market over the same period. Over the five years through early 2026, VNQ gained roughly 2.6 percent cumulatively while the S&P 500 rose about 85 percent, a gap driven largely by the Federal Reserve's rate-hiking campaign that began in 2022: REITs typically carry significant debt, so higher borrowing costs compress their valuations and make their dividend yields less competitive against rising bond yields.

A recovery thesis for REITs generally depends on falling interest rates, and as of early June 2026 that case looked more favorable, with expectations for Federal Reserve rate cuts later in the year. That picture shifted with the inflation reading released June 10, 2026: annual inflation accelerated to 4.2 percent, the third consecutive monthly increase and the highest level since April 2023, driven substantially by an energy price shock. A Federal Reserve facing accelerating headline inflation has historically been reluctant to cut rates regardless of the underlying cause, which weakens, without eliminating, the rate-cut-driven case for a REIT recovery in the near term.

Even setting aside interest rates, a broad fund like VNQ dilutes any data center or AI theme substantially. Equinix, Digital Realty, and Iron Mountain represent a modest fraction of VNQ's holdings, the bulk of the fund is apartments, offices, malls, healthcare facilities, and other property types with no relationship to AI infrastructure at all. An investor buying VNQ for AI exposure is, in effect, two steps removed: once from the REIT to the hyperscaler tenant, and again from the data center REITs to the broader real estate basket.

Where the AI Profit Actually Sits

The Honest Conclusion

Data center REITs are real businesses benefiting from real AI-driven demand, the scarcity of power and land that REITs control has genuine value, and the recent performance of Equinix, Digital Realty, and Iron Mountain reflects that. But for an investor whose specific goal is exposure to AI's profitability, REITs occupy the landlord's position in a value chain where the tenant captures most of the upside. A technology-focused fund such as Vanguard's VGT, or direct positions in the hyperscalers and chip suppliers, captures AI-driven returns more directly than any REIT structure can, simply because that is where the AI revenue itself is recognized.

This does not make REITs a poor investment in general, dividend income, lower volatility than AI-concentrated tech positions, and diversification away from a narrative that, as this publication has discussed with Tesla and SpaceX, can carry its own fame-driven premium, are legitimate reasons to hold them. But those are reasons to hold REITs as REITs, for what real estate has historically offered a portfolio, rather than as a way to participate in the AI buildout itself. The clearest takeaway is simply to be precise about which thesis is being expressed: "AI infrastructure is being built at enormous scale" is true and supports data center REITs to a degree, but "I want exposure to AI's profits" points toward the companies generating those profits, not their landlords.

Figures reflect company filings, REIT performance data, and Bureau of Labor Statistics inflation data current as of mid-June 2026, and are subject to change.