AI Growth, Data Centers Reshape Real Estate

Data centers are no longer a niche bet in commercial real estate: they’re on track to become the biggest construction story in the business and a growing source of both upside and anxiety for property investors.

Spending on U.S. data center construction is set to surpass office-building construction as soon as next year, according to U.S. Census Bureau data cited by the Wall Street Journal. It’s a symbolic milestone that underscores how deeply artificial intelligence has rewired the real estate capital stack.

The money has been hard to ignore. Data centers returned 11.2 percent last year, outperforming every major property type except manufactured housing, according to the National Council of Real Estate Investment Fiduciaries. At the same time, office construction has stalled amid record vacancies and a multiyear glut that shows little sign of easing.

Data center demand is being driven by hyperscalers racing to build out AI infrastructure.

Meta, Amazon, Oracle and others are increasingly leasing capacity from landlords instead of building their own campuses. Roughly 40 percent of U.S. hyperscaler data center capacity was leased this year, up from about 35 percent in 2023, according to McKinsey estimates.

That shift has pulled institutional capital deeper into the sector. JLL projects as much as $1 trillion in data center construction across North America between 2025 and 2030.

Publicly traded real estate investment trusts boosted their data center exposure by 15 percent last year while cutting back on offices and apartments, according to Nareit. A CBRE survey found 95 percent of major investors plan to increase allocations to the sector.

The result is a structural change in how real estate behaves in a downturn. Historically, property portfolios offered insulation from tech cycles. During the early-2000s dot-com crash, commercial real estate values held relatively steady even as the Nasdaq cratered.

Data centers blur that line. They are highly specialized assets tied to a narrow tenant base and, increasingly, to the fortunes of AI itself.

Long-term leases — often 15 to 20 years — provide comfort and tenants like Oracle have the balance sheets to backstop commitments. But those agreements come with complex clauses tied to power delivery, construction deadlines and uptime that can expose landlords to steep penalties or cancellations.

Add in labor shortages, supply-chain risks and grid constraints, and the boom starts to look less frictionless. Data centers may be replacing offices as real estate’s growth engine, but they’re also binding the industry more tightly than ever to the trajectory of a still-unproven AI economy.

— Holden Walter-Warner