Data-Center Real Estate

The question

The question for data-center real estate is narrower than the popular framing suggests. It is not whether hyperscalers will take down more capacity — they will — but whether the market has already capitalized the lease-up assumptions at a multiple that leaves adequate margin of safety. The pre-screen filter we use — one-year total return relative to SPY — is a blunt instrument, and in this cohort it disqualifies most of the obvious names. What remains is a concentrated book of seven holdings where contracted cashflow has kept pace with, or outpaced, the share-price rerating. We acknowledge the concentration up front: this is the survivors-of-the-rerating cycle, not the full universe.

Base rates

The base rate for specialty REIT outperformance during a demand inflection is more favorable than the generalist real estate base rate, but the distribution inside the specialty cohort is wide. Studies of tower REITs (2005-2015) and self-storage REITs (2010-2018) show median outperformance of 400-600bps annualized during their respective demand inflections, but with top-quartile names delivering 1,200bps+ and bottom-quartile names delivering negative relative returns. The dispersion is explained almost entirely by two variables: cost-of-capital advantage at the start of the cycle, and the ability to pre-lease development pipeline at returns above the cost of incremental debt. Our reference class is the top quartile of the tower-REIT analogue, and the names in this book either already clear that bar (EQIX, DLR) or are proxies for adjacent physical infrastructure where the same dynamics apply (WELL, PLD, IRM, EME, UNIT).

Why the consensus view is wrong (or incomplete)

The consensus reads data-center REITs as a direct play on hyperscaler demand. That reading is correct but incomplete in two ways. First, it under-weights the role of power availability as the binding constraint on supply; the names with secured power and interconnection in constrained markets have pricing power that is materially higher than the headline same-store rent roll suggests. Lease-up on new product is happening at 30-50% higher rents per kilowatt than lease rollovers on existing product, and the market is blending these into a single growth rate. Second, the consensus conflates the wholesale (hyperscaler) and retail (enterprise colocation) segments. The wholesale segment is rerating for demand reasons; the retail segment is rerating for a different reason — enterprise AI adoption requires low-latency colocation in specific metros, and the pricing in those metros has moved independently of hyperscaler leasing. EQIX is the cleanest expression of the retail dynamic; DLR captures both.

Position construction

The book carries only seven holdings because the 1Y-vs-SPY filter is strict and most of the data-center REIT cohort has underperformed SPY in the last twelve months despite strong operating results. That underperformance is itself informative: the cohort already rerated in 2023-2024, and the survivors of our filter are the names whose cashflow has caught up to the multiple rather than the names still trading on narrative. We group the holdings into two sub-books.

Core real estate franchises (80%) — PLD 20%, EQIX 20%, DLR 19.3%, WELL 20%. PLD is included because its power-and-land-bank optionality has been undervalued; its development pipeline in the Northern Virginia, Phoenix, and Chicago markets overlaps directly with hyperscaler demand. EQIX is the retail colocation leader with pricing power anchored in interconnection density. DLR is the hybrid — wholesale scale with a retail colocation business attached, and the cleanest free-cash-flow inflection in the cohort. WELL is the non-obvious inclusion: its senior housing franchise has a demographic tailwind that is independent of the AI thesis, and it passed the 1Y-vs-SPY filter with room to spare, so it sits in the book as a non-correlated source of contracted cashflow.

Installer and fiber tail (20%) — EME 10.1%, IRM 10.1%, UNIT 0.5%. EME is the mechanical-electrical-plumbing contractor that builds the power and cooling infrastructure inside the data centers themselves; its backlog extends past 2028 at pricing that reflects post-2024 scarcity. IRM is the storage-to-data-center conversion story, where the market is still capitalizing its data-center segment at a discount to pure-play comparables. UNIT is a small residual position in fiber real estate; it passed the filter but we are not sizing it as a conviction name.

Asymmetric payoff

Our expected-value frame: base case (50%) delivers 10-12% annualized on stabilized FFO growth of 7-9% plus dividend; upside (25%) delivers 18-22% if lease-up on development pipelines accelerates and cap rates compress another 50-75bps; downside (25%) delivers flat-to-negative mid-single-digits if hyperscaler capex is cut and development product delivers into a softer market. Probability-weighted, the book offers low-teens expected return with an asymmetry that favors the upside because the contracted-revenue floor is more defensible than in the sponsor or power-generation books.

Three things that would change our mind

First, a move in long rates above the 2023 highs that would compress cap rates across the cohort before FFO growth can offset it. Second, a meaningful shift in hyperscaler strategy toward owning rather than leasing data-center capacity; leased capacity is currently 55-60% of hyperscaler footprint and any move below 45% would change the demand mix. Third, a supply response from new entrants (sovereign wealth-funded developers, private equity rolling up secondary markets) large enough to compress the lease-up spread we are underwriting.

What we are explicitly NOT betting on

We are not betting on a particular hyperscaler's capex trajectory. We are not betting on the retail colocation cycle decoupling fully from the wholesale cycle. We are not betting on a cap-rate compression driven by Fed cuts; the book should work at flat rates. And we are not betting on UNIT as a turnaround; it is a residual position, not a thesis.