Gail Parcell does not exist, which is a shame, because the data center industry could use her. She is a composite, assembled from the professional reality of infrastructure site selection consultants whose daily work involves reading geological and bureaucratic strata that most people never think about. We spoke with her over a video call from what she described as "a Holiday Inn in central Iowa that has better fiber connectivity than most office buildings in San Francisco." She did not explain how she'd tested this.

You evaluate sites for data centers. What does that actually mean on a Tuesday morning?

Gail: It means I'm standing in a field, or a parking lot, or sometimes what used to be a manufacturing plant, and I'm trying to figure out what's underneath it. Not geologically. Infrastructurally. What conduit runs through here. What power feeds this parcel. What rights-of-way were cleared, and when, and by whom. The thing sitting on top of the land is almost never the point. I've evaluated beautiful greenfield sites with great tax incentives that are essentially worthless. And I've evaluated ugly parcels next to railroad tracks worth ten times what the owner thinks, because there's a fiber trunk line six feet below the gravel.

When did you start seeing that pattern?

Gail: It took me embarrassingly long. I came into this from commercial real estate in the late '90s, and I kept applying the same logic: location, square footage, proximity to workforce. None of that is wrong, exactly, but it's maybe fourth or fifth on the list of what actually determines whether a site works. The first time I really understood was when a client wanted to build in the Midwest and I kept pulling up these rural parcels along the Iowa-Nebraska border with extraordinary fiber connectivity. Made no sense until someone handed me a railroad map from the 1860s.¹

What did the railroad map explain?

Gail: Everything. The transcontinental railroad defined the corridor. Telegraph lines went up alongside the tracks, same right-of-way, same stations.² Telephone lines followed the telegraph. Then in the '80s and '90s, AT&T and MCI paid railroads for easement rights to plow fiber optic cable alongside those same tracks.³ They laid far more fiber than they needed because the marginal cost of additional strands was almost nothing once you had the trench open. Most of it sat dark for years. And now that dark fiber is getting lit up for AI training clusters. A cornfield in Iowa with a transcontinental fiber trunk underneath it will beat a lot of East Coast industrial parks on a dollar-per-megawatt basis. I can show you the spreadsheet. It's not even close.

So the fiber follows the railroad, which follows what? The terrain?

Gail: River valleys, mountain passes, whatever the surveyors found in the 1860s. And here's the thing that trips up every new developer I work with: fiber does not follow straight lines between cities. It follows rights-of-way.⁴ A site that looks 40 miles from a major hub might actually be sitting on 90 miles of fiber because the route follows the old rail corridor around a ridge. Or the reverse. A site that looks remote on a map might be directly on a transcontinental route because the railroad went through there 150 years ago.

You cannot evaluate a data center site using Google Maps. You need GIS mapping of actual fiber paths.⁵ I've watched people spend six months on a site before discovering this.

What about the buildings themselves? I've heard former telephone switching stations are particularly valuable.

Gail: They're gold. A telephone central office from the mid-20th century was engineered for exactly the conditions a data center needs. Heavy floor loads, because switching equipment weighed a ton, literally. Massive power feeds, often transmission-level, not just distribution. Cooling systems designed for continuous heat-generating equipment. And most importantly, conduit infrastructure. The pathways bringing cable into that building were permitted, cleared, and maintained for decades. Multiple carriers already have fiber terminating there. You cannot replicate that in a new build without years of permitting and construction.

There's a developer in South Bend, Indiana, building a data center hub in an old central train station because of the fiber running adjacent to the railways.⁶ People hear that and think it's charming. It's the most rational economic decision available to them.

You mentioned power. That seems to be the constraint everyone's talking about.

Gail: It's the constraint. A site with 100 megawatts would've been a big project in 2022. Now we're looking at campuses designed for over a gigawatt.⁷ The grid can't keep up. Utilities are telling developers five, seven, ten years for new transmission capacity. So what wins? Sites that already have heavy electrical infrastructure. Former manufacturing plants. Military bases. Telecom switching hubs. Places where someone in 1950 or 1970 decided to run serious power, and that infrastructure is still there, still connected, still permitted. Amazon's been pursuing sites adjacent to nuclear power stations for direct power purchase.⁷ That tells you everything about how constrained things are.

It sounds like the AI boom is being shaped by decisions made before anyone imagined it.

Gail: That's what I keep trying to explain to people who think this is about finding cheap land with good tax incentives. The land is almost irrelevant. What matters is what previous generations built on it, under it, and through it.

Every generation installs infrastructure for its own purposes, and every subsequent generation discovers it can't escape the geometry of what came before.

Does that bother you?

Gail: Why would it bother me? It's my entire career. If infrastructure were rational, if you could just pick the optimal point on a map and build, I'd be out of a job. The fact that you need someone who can read a railroad map from 1869 to evaluate an AI data center site in 2026? The decisions compound. The corridors persist.

Footnotes

1. Major data center campuses from Google, Microsoft, and Meta are concentrated along the Iowa-Nebraska border, coinciding with the transcontinental railroad corridor. See: https://www.atlasobscura.com/articles/mapping-the-hidden-structures-of-new-york-citys-internet-networks ↩

2. Telegraph operators preferentially shared railway corridors from the earliest U.S. telecommunications buildouts. World Bank Cross-Sector Infrastructure Sharing Toolkit (2017): https://thedocs.worldbank.org/en/doc/307251492818674685-0190022017/render/CrossSectorInfrastructureSharingToolkitfinal170228.txt ↩

3. AT&T and MCI paid railroads for easement rights in the 1980s to install fiber alongside tracks, laying far more capacity than needed. See: http://industrialscenery.blogspot.com/2017/09/fiber-optic-plow-for-trackside-burial.html ↩

4. "Don't assume geographic straight lines represent actual fiber paths. Fiber follows rights-of-way, which means routes might detour substantially from direct paths." Netrality (2025): https://netrality.com/blog/data-center-interconnect-guide/ ↩

5. GIS mapping of actual fiber routes is the operative tool in data center site evaluation. SelectROW (2024): https://selectrow.com/data-center-site-selection/ ↩

6. South Bend train station data center development cited in Atlas Obscura: https://www.atlasobscura.com/articles/mapping-the-hidden-structures-of-new-york-citys-internet-networks ↩

7. Power demand, gigawatt-scale campuses, and nuclear-adjacent siting from Procore (2025): https://www.procore.com/library/data-center-site-selection ↩ ↩²