AI Data Centers: Why The Landlords Are Lagging-and How To Catch Up
AI demand is swallowing every megawatt it can find. Chips and cloud platforms are racking up trillion-dollar wins, yet the largest data center landlords-Equinix, Digital Realty, and Iron Mountain-have traded down over the last year while the S&P 500 climbed.
On paper, this should be their moment. In practice, the market thinks the margins sit with the chips and the hyperscalers, not with the shells and switches.
What's holding the REITs back
First, the structure. REITs must pay out 90% of taxable income as dividends. That caps how much they can reinvest and makes it harder to pre-commit massive sums for multi-gigawatt campuses that AI customers now expect.
Second, risk tolerance and leverage. Public data center REITs typically run below 5x leverage. Private operators and infra-backed developers often run 10-15x, pair with aggressive capital partners, and accept customer credit profiles with hair on them. REIT investors-often pension-style-don't love that playbook.
Third, contract and delivery risk. Many hyperscale deals include early termination rights tied to delays. More than half of projects in 2025 slipped by three months or more. Pair that with utility bottlenecks, and empty shells without energized megawatts drag returns.
The market read: chips over sheds
Analysts argue that Google, Nvidia, Broadcom, and friends are positioned to capture the bulk of AI economics. That view hardened as the biggest 2025 development mandates skewed to private developers and owner-builders instead of REITs.
Stock reaction tells the story. Oracle flagged heavier data center spend after a soft quarter and popped. Equinix signaled heavier AI capex alongside lower near-term revenue and sank. Same theme, different risk profiles.
Location strategy: core metros vs remote land
REITs have stuck to Northern Virginia, Los Angeles, and other core hubs. Private players and hyperscalers quietly bought land where power, water, and setbacks are easier-farther out, bigger footprints.
That said, there's an opening. As workloads shift from model training to inference, latency matters. Urban and near-urban interconnection hubs-with rich fiber and proximity to users-gain urgency for running models at scale.
What real estate and construction teams can do now
Capital and structure
- Stand up development JVs with infra funds or sovereigns for off-balance sheet firepower. Use pref equity and asset recycling: sell stabilized interconnection assets, redeploy into AI build-to-suits.
- Create a development "opco" paired with REIT holdco via ground leases or taxable REIT subsidiaries. This lets you take calculated construction risk without contaminating dividend stability.
- Stage capital commitments with milestone-based draws tied to utility milestones and long-lead equipment receipts.
Power first, dirt second
- Secure queue positions early and in bulk. Lock transformer and switchgear orders up front; these are your critical path.
- Pursue dual-track power: utility interconnection plus firmed PPAs with storage or behind-the-meter generation where allowed.
- Co-fund dedicated substations with utilities; bake lead times into term sheets. Consider options on adjacent parcels for future feeders.
If you need a primer on the grid bottleneck story, this is a useful overview from the U.S. national labs: interconnection queue analysis (Berkeley Lab).
Design for density and speed
- Standardize a high-density, liquid-ready kit: hot aisle containment, CDU rooms, rear-door HX or direct-to-chip readiness, and 30-80 kW/rack paths.
- Use modular plant rooms and phased energization. Turn on power in blocks to start revenue while the rest builds out.
- Engineer for water stewardship from day one: recycled sources, adiabatic-free options where water constrained, and heat reuse where practical.
De-risk the contract
- Replace broad early termination with targeted liquidated damages and step-in rights. Tie schedule risk to the actual bottlenecks: utility energization and long-lead equipment.
- Require parent guarantees or LCs for newer credits; use ring-fenced SPVs to isolate counterparty and project risk.
- Index pricing to power density and input costs with collars. Take prepayments to fund long-lead gear.
Two-speed location thesis
- Training: remote or exurban gigacampuses with abundant land and easier permitting. Focus on cheap, firm power and water.
- Inference: metro-adjacent nodes with deep fiber and low-latency routes to users and enterprise data. This is where Equinix and similar interconnect fabrics keep an edge.
For context on market hotspots and delivery challenges, see the latest JLL data center outlook.
Signals to watch
Equinix is scaling its xScale program with cloud majors-proof it can pivot to larger AI blocks while protecting its interconnection moat. Digital Realty's AI workload mix and pre-lease cadence will show whether hyperscale shifts back their way. Iron Mountain's credit mix and leverage tolerance will tell you how far public data center balance sheets are willing to stretch.
Meta's stated plan for tens of gigawatts this decade (and more over time) implies trillions in spend at $50B per GW. That tide lifts many boats-but only the ones with power, parts, and partners in place.
Quick checklist for your next AI build
- Utility: queue position filed, substation co-funding mapped, dual-feed and N-1 scenarios priced.
- Procurement: transformers, switchgear, generators, CDUs, and cooling hardware reserved with penalties for slips.
- Design: liquid-ready, high-density, modular blocks with phased energization.
- Contracts: LDs over termination, step-in rights, indexed rent, prepayments, and parent guarantees.
- Capital: JV or pref equity lined up, asset recycling plan, and a clear REIT/TRS structure for development risk.
- Siting: split strategy-remote for training, metro-adjacent for inference-with fiber routes verified.
Level up your team
Preconstruction, PM, and MEP teams that speak AI workloads move faster and win the better credits. If you need structured upskilling, here's a practical place to start: AI courses by job role.
The takeaway: there's more than enough demand. The winners will be the developers who secure power early, standardize high-density delivery, shift risk with smart contracts, and bring flexible capital to the table. Move first on those four, and the market will re-rate you.
Your membership also unlocks:
