These ‘underappreciated winners of AI’ also pay solid income

Data center REITs are highlighted as underappreciated AI beneficiaries, with the subsector up nearly 40% year to date as of April 30. Equinix yields 1.9% and has nearly 14% upside to the average target, Digital Realty yields 2.5% with almost 16% upside, and Iron Mountain yields 2.7% with about 5% upside. The article frames the group as a durable, income-producing way to gain AI exposure, supported by strong analyst sentiment and continued demand for data center capacity.

Analysis

The market is correctly pricing the AI buildout as a compute story, but the cleaner economics may sit in the landlord layer: data-center REITs monetize scarcity through long-duration contracts, power access, and interconnection density, which are harder to replicate than GPUs or software. That creates a second-order beneficiary set beyond the obvious three names: electrical equipment, grid interconnection, cooling, and even specialized debt providers should see pull-through as tenants rush to secure capacity and developers need external capital.

The key nuance is that the next leg of demand is likely inference, not training. Inference is more distributed, latency-sensitive, and sticky to population centers and network hubs, which structurally favors incumbents with dense metro footprints and carrier ecosystems rather than greenfield hyperscale campuses. That should widen the gap between “real estate with power” and commodity colocation over the next 12-24 months, especially if power availability becomes the binding constraint rather than land.

The main risk is that consensus is extrapolating pricing power without enough attention to capacity additions already in the pipeline. If new supply, utility approvals, or financing conditions loosen, rent growth could normalize faster than share prices imply, particularly for the more straightforward operators. Another risk is duration: the equity story benefits from long-term AI adoption, but multiples are already discounting several years of strong NOI growth, so disappointment can show up quickly if absorption lags for even two quarters.

The contrarian view is that the better risk-adjusted trade may not be the broad basket but the highest-quality networked platform with the strongest balance sheet, since it can compound through both rental escalation and development optionality. The crowded part of the trade is the pure AI narrative; the less crowded angle is the income stream plus capital recycling advantage of public REIT status versus private owners. If the public market keeps rewarding access to debt capital, listed operators should gradually gain share from private portfolios that cannot scale as efficiently.

Key Decisions for Investors

• Overweight EQIX vs DLR on a 6-12 month horizon: EQIX has the cleaner network-density moat and should capture a larger share of inference-driven interconnection demand; target relative outperformance of 5-8%, but trim if the pair re-rates beyond a 2.0x EV/EBITDA premium to DLR.
• Long DLR on pullbacks over the next 1-3 months: use it as the higher-beta way to express continued supply-demand tightness; upside is attractive if capacity remains constrained, but stop if occupancy or lease spreads roll over for two consecutive quarters.
• Hold/accumulate IRM only as a barbell income trade, not as the core AI exposure: the rerating has already compressed the margin of safety, so treat it as a yield vehicle with limited upside unless management proves materially better AI-linked development economics.
• Initiate a basket long EQIX/DLR vs short a broad real estate index over 3-6 months: this isolates the AI-capex beneficiary set from rate-sensitive commercial real estate, with better earnings momentum and less dependence on macro growth.
• Watch for a secondary beneficiary trade in power and cooling infrastructure suppliers over 6-18 months; if data-center REIT capex stays elevated, the next incremental winners are names with utility interconnect, switchgear, and liquid-cooling exposure rather than the REITs themselves.