Data centers are the new trophy office (and the new risks)

Per JLL's year-end 2025 report, data center vacancy hit 1% in 2024 and held there through 2025. Rents rose 9% in 2025; renewals on large deployments commanded 13% increases. The five largest hyperscalers (Amazon, Microsoft, Google, Meta, Oracle) announced $710B of planned 2026 capex — enough for 35 GW of new global capacity. US capacity is projected to rise from ~30 GW in 2025 to 90+ GW by 2030 (22% CAGR).

The constraint has moved from capital to power. Grid connection timelines now average 4+ years in major hubs. PJM's 2025/2026 capacity auction priced at $14.7B vs. $2.2B the prior year. ERCOT was monitoring 233+ GW of large load interconnection requests as of December 2025 — up almost 300% over 2024 year-end. Power transformer lead times reached 128–144 weeks per Wood Mackenzie. For the asset owner with secured power and interconnection, the constraint is a moat.

For individual investors, Digital Realty (DLR) and Equinix (EQIX) are the two major public REIT pure-plays. Both trade at premium P/FFO multiples — 25x+ versus mid-teens for most other REITs. The premium is justified by the structural growth but creates downside if AI capex pulls back.

The concentration risk worth naming: ~92% of data center capacity under construction is pre-committed to a concentrated set of roughly 10 hyperscalers. If those 10 customers materially slow their AI capex (a 30%+ pullback in the bear case), the entire vacancy thesis could re-rate quickly. This is similar to how trophy office concentrated in 5–10 major tenant industries (finance, tech, consulting) — and we know how that ended for commodity office.

Family-office angle: direct equity in vertically integrated data center operators (Vantage, CyrusOne, smaller hyperscale build-to-suit specialists) is the highest expected-return position in the category for capital able to take operating risk. Pure capital allocation through REITs gets you the beta; operating-platform stakes get you the alpha.