A $5 million construction estimate for a 226,469-square-foot data center shell is not the number that matters. The number that matters is the $165 million Starwood paid for the 34 acres underneath it in 2022. That land cost, roughly $4.85 million per acre, is the real underwriting variable. The shell is cheap. The dirt is the bet.
Starwood Capital Group filed plans last week for Plaza 500 South, the first of two data centers it intends to build on the site near the Alexandria border in Fairfax County. The two-story building would rise 75 feet. The first-phase shell cost is estimated at $5 million. That is a rounding error in data center development. The real capital commitment is the land, the power infrastructure, and the time required to make both productive.
Starwood acquired the site in 2022 for $165 million. That was peak pricing for Northern Virginia land, when data center demand was accelerating and every institutional sponsor was racing to secure sites near the world's largest internet exchange points. Three years later, the land is still vacant. The shell has not been built. And the project now depends on a proposed electrical substation that would sit less than 100 feet from nearby homes in the Bren Mar neighborhood.
That substation is the project's gating item. Without it, the data center cannot operate. With it, the project faces organized resident opposition, a planning commission hearing in September, and the kind of public-process uncertainty that lenders typically price as a binary risk: either the substation gets approved on a predictable timeline, or it does not. There is no middle ground.
Residents have already proposed a land swap: Fairfax County would take the Plaza 500 site for affordable housing and sell Starwood a different county-owned site in Chantilly for data center development. County officials rejected the idea and are expected to close on the $167 million Chantilly property sale to Starwood next spring. That second transaction is relevant because it shows the county is willing to sell Starwood land at a comparable price. It also shows that Starwood is willing to buy two sites simultaneously, which means the firm is placing a large, concentrated bet on Northern Virginia power availability and zoning outcomes.
A credit committee reviewing a construction loan for Plaza 500 South would have to answer a question that has nothing to do with tenant credit or lease term. The question is: how do you underwrite a project whose economic viability depends on a substation that has not been approved, that faces organized opposition, and that sits 100 feet from residential property lines?
The answer, for most bank credit committees, is that you do not. Construction lenders have become deeply cautious about data center projects that require new utility infrastructure, especially when that infrastructure must be built in or near residential areas. The timeline risk is too long. The public-process risk is too binary. And the capital at stake is too large relative to the fee income a construction loan generates.
This is why data center development in Northern Virginia has increasingly shifted toward sponsors with the balance sheet to self-fund the infrastructure timeline. Starwood, with roughly $145 billion in assets under management, has that capacity. But even for a firm of that scale, carrying $165 million in land for three years without income is a real cost of carry. At a conservative 5 percent weighted average cost of capital, that land has cost roughly $25 million in carrying costs since acquisition. The $5 million shell estimate is trivial by comparison.
The project's economics therefore depend on two things: the speed of the substation approval and the eventual lease rate. Amazon is already slated to use space at Starwood's nearby Renaissance Tech Park in Herndon, which is expected to span 2.3 million square feet at full buildout. That relationship suggests Starwood has a credible path to tenant demand. But a lease at Renaissance Tech Park does not guarantee a lease at Plaza 500. Each site has its own power timeline, its own construction schedule, and its own approval risk.
The market signal here is not about data center demand. That demand is well established. The signal is about the cost and complexity of bringing new supply online in the world's most competitive data center market. Northern Virginia has the lowest vacancy rates and the highest absorption in the country. It also has the longest lead times for power delivery and the most organized community opposition to new substations and transmission lines.
For lenders, the implication is straightforward: data center construction loans are no longer pure real estate credit. They are infrastructure credit with a real estate wrapper. The underwriting must account for utility approval timelines, community opposition risk, and the sponsor's willingness to carry land through multi-year approval processes. The loan structure must include milestones tied to substation approval, not just building completion. And the pricing must reflect the possibility that the substation never gets built.
For owners and developers, the lesson is equally direct. Land basis matters more than ever. Starwood paid peak pricing in 2022. That basis is now carrying cost that compounds every quarter the substation remains unapproved. The project can still work, but only if the approval timeline compresses and the lease rate justifies the total capital stack. If the substation takes another two years, the land cost alone will have consumed a material portion of the project's expected return.
The September planning commission hearing will be the first real test. If the substation is approved, the project moves forward with a known timeline and a known cost. If it is delayed, the land carrying cost continues to compound, and the basis becomes harder to defend. A credit committee watching from the sidelines will learn more from that hearing than from any lease proposal.