Artificial intelligence has pushed data centers to the center of the commercial real estate conversation. Capital continues flowing into digital infrastructure as cloud providers and technology firms race to expand computing capacity. Development pipelines are growing across major markets, and investors increasingly treat the sector as one of the most attractive areas of institutional real estate.
Yet the structure of the business is starting to shift.
Rising development costs and changes in leasing strategy are altering how these assets are financed and underwritten. The combination may reshape the way investors evaluate data center real estate over the next phase of the cycle.
Development Requires Capital Much Earlier
Building large-scale data centers now requires substantial capital commitments well before construction begins.
Securing land and power has become the first major hurdle. According to a recent Bisnow report, prices for U.S. data center development sites larger than 50 acres climbed more than 23 percent last year. The same report noted that utilities in some markets now require commitments exceeding $400 million before power infrastructure can be delivered.
Those requirements push developers into a capital-intensive predevelopment phase that did not previously exist at this scale. Land acquisition, transmission upgrades and power allocation must often be secured years before a project reaches stabilization.
Traditional project finance models struggle with that timeline. Many lenders prefer signed tenant commitments before deploying large amounts of capital, while tenants typically require confirmed infrastructure capacity before executing long-term leases.
The mismatch has accelerated the use of alternative funding structures. Developers increasingly recycle capital by selling stabilized facilities, raising dedicated development funds or structuring securitizations backed by operating assets in order to fund the next generation of projects.
Leasing Structures Are Starting to Change
The leasing model that historically defined the sector is evolving as well. Data centers were once associated with long-duration leases that resembled infrastructure contracts. Tenants frequently signed agreements lasting ten to fifteen years, creating predictable cash flow streams that institutional investors valued.
Artificial intelligence workloads are beginning to alter that pattern. A recent Moody’s analysis found that major technology companies are increasingly negotiating shorter lease terms aligned with the rapid hardware refresh cycles that characterize AI computing infrastructure.
Servers and AI accelerators can become obsolete in only a few years. As a result, tenants increasingly favor agreements that allow facilities to adapt to new equipment generations without locking the company into long-term physical constraints.
That shift does not eliminate tenant credit strength, but it does compress the timeline of income visibility that many investors have traditionally relied on when underwriting data center assets.
The Scale of Commitments Is Enormous
The amount of capital tied to this expansion is also larger than many observers realize. Moody’s estimates that the largest technology companies, including Amazon, Microsoft, Google, Meta and Oracle, have collectively committed hundreds of billions of dollars to future data center leases tied to AI infrastructure. Many of those commitments do not appear on company balance sheets until the lease periods officially begin, according to the ratings agency.
The accounting treatment means the financial exposure associated with digital infrastructure expansion can remain partially hidden during the development phase. As projects come online and leases commence, those obligations gradually move onto corporate financial statements.
The result is a clearer picture of how deeply the AI economy depends on physical real estate.
A Different Kind of Infrastructure Asset
Data centers have often been described as the closest thing commercial real estate has to infrastructure. Long leases, investment-grade tenants and steady demand historically made the sector appear predictable relative to other property types.
The AI cycle introduces new variables.
Development now requires significant capital before tenants are secured, while leasing structures increasingly reflect the pace of technology upgrades rather than the lifespan of buildings. Those dynamics do not reduce demand for digital infrastructure, but they do alter how these assets are financed, underwritten and valued. This will in turn reshape investment sales in countless markets where these data centers currently exist or are needed in the near future.
