Kind: Measured report
By Asif Hussain, Founder and CEO, SpecIndex
Since January 2025, SpecIndex's own database of public bid and permit records has logged 349 data center projects across the United States, with a combined estimated value of $226 billion. Virginia alone accounts for $77.9 billion of that figure, spread across just 39 projects, an average project size that dwarfs every other state in the pipeline.
That concentration is the story. Data center construction is not spreading evenly across the map the way warehouse or multifamily construction does. It is piling up in a small number of counties that already have the two things a hyperscale campus needs before it needs anything else: substation capacity and fiber.
Facts at a glance
- Since January 2025, SpecIndex's own database of public bid and permit records has logged 349 data center projects across the United States, with a combined estimated value of $226 billion. (From: the opening paragraph.)
- Virginia alone accounts for $77.9 billion of that figure, spread across just 39 projects, an average project size that dwarfs every other state in the pipeline. (From: the opening paragraph.)
- The $226 billion figure is a floor built from what is on the public record since January 2025, not a market wide estimate. (From: Method note and scope.)
Virginia wins on dollars, and Texas wins on count
Ranked by project count, the states in the table below lead the pipeline in both projects and dollars in SpecIndex's corpus since the start of 2025.
| State | Projects (2025 to present) | Estimated value |
|---|---|---|
| Texas | 79 | $8.6 billion |
| Georgia | 46 | $1.0 billion |
| Virginia | 39 | $77.9 billion |
| California | 28 | $639 million |
| Illinois | 25 | $3.5 billion |
| New York | 22 | $20 million |
| Arizona | 16 | $3.6 billion |
| Wyoming | 5 | $9.2 billion |
Download CSV (the rows above, exactly as shown)
*Table note: New York's $20 million is the value carried in the records counted here for its 22 projects. Read state value totals as floors, on the same basis as the national floor stated in the method note, and not as market totals for a state.*
Two things jump out at once, and they are the same two things any analyst covering this sector runs into.
Virginia wins on dollars, Texas wins on count. Virginia's Data Center Alley in Loudoun and Prince William counties has the oldest and densest hyperscale infrastructure in the country, so new projects there tend to be large expansions of existing campuses rather than greenfield builds.
That is why its 39 projects carry the nine and ten figure valuations that push the state's total past every other market combined. Texas, by contrast, is in an earlier and more distributed phase: 79 separate permitted projects, a lower average size, and activity spread across multiple metros rather than one corridor.
Wyoming is a genuine outlier, not noise. Five projects and $9.2 billion in value means an average project size north of $1.8 billion, larger even than Virginia's average. That is consistent with the large hyperscale campuses announced for the state's low cost power and land in the last two years, and it is a reminder that a state's rank in this pipeline has almost nothing to do with its size or its prior data center history.
Power availability gates the map, not population
Three drivers explain the state level variation SpecIndex's records show, and power comes first among them.
Power availability, not population, is the gating factor. A hyperscale campus needs a substation scale interconnection, and a single campus phase can need 100 to 300 megawatts.
Utilities in Virginia, Georgia and Texas have made large transmission investments specifically to serve this demand. States without that headroom are not in the pipeline regardless of land cost or tax incentive.
Tax abatement drives the marginal decision, not the first one. Every state at the top of the table offers some form of sales or property tax abatement for qualifying data center investment.
That incentive determines which of several power capable sites a developer picks. It does not create demand where the power does not exist.
Fiber backbone proximity compounds the advantage. Northern Virginia's dominance predates the current boom by two decades, because the original internet exchange infrastructure concentrated there. New capacity keeps landing near existing capacity because the fiber routes and the skilled trade labor pool are already there.
The filing sequence tells you more than the press release
A data center campus generates a distinctive sequence of public filings, and reading that sequence is a more reliable early indicator than press releases, which can use a code name for the project rather than the operator's own name.
The earliest public signal is a grading or site development permit, followed months later by a building shell permit, and then the filing that actually confirms hyperscale intent: a series of large electrical service and generator permits under Division 26 (Electrical) and Division 23 (HVAC) for the cooling plant. A single tenant enterprise data center rarely generates the generator and switchgear permit volume that a hyperscale campus does. The mechanical and electrical filing pattern is a better tell than the building permit itself.
This is also where SpecIndex's specification level data adds something a project tracking service cannot. The mechanical and electrical divisions on a hyperscale data center campus are where the real manufacturer competition happens: chillers, generators, switchgear and cooling distribution units, specified section by section and campus by campus, well before the equipment is ordered. A developer or design team running the same campus format across multiple states tends to reuse the same basis of design manufacturers from site to site, which means the first state's specification can predict the next one's.
Method note and scope
We counted data center projects in SpecIndex's own database of public bid and permit records. The population is the 349 data center projects logged in that corpus since January 2025, and the match rule is the one the article itself describes: a project is in the count when the corpus records it as a data center project that passed through a public permitting authority, with every project in the table above having cleared a county or state building department. The window is January 1, 2025 onward.
The figures in this article, including the 349 projects, the $226 billion combined estimated value, Virginia's $77.9 billion across 39 projects, and the state table, are retained from the original publication and are dated as of the article's original September 2026 count. The corpus keeps growing, so the figures are dated and later counts will differ. New York's line belongs on that same footing: its $20 million across 22 projects is the value in the records counted here, and the method note treats state value totals as floors rather than market totals.
Scope has limits. The corpus will undercount activity in jurisdictions with slower or less digitized permit publication, and it cannot see a project before its first public filing, which for a hyperscale campus can be a year or more after the real estate and power agreements are signed. The $226 billion figure is a floor built from what is on the public record since January 2025, not a market wide estimate.