Infrastructure
The Buildout
Satellite imagery reveals how AI data centers are reshaping the American landscape — from factory floors to server halls, farmland to construction sites — in months, not years.
Midlothian, Texas. A quiet stretch of open land south of Dallas. Drag the slider to see what replaced it — a sprawling Google data center campus that now draws more electricity than a small city. Across the country, the same transformation is playing out on farmland, industrial parks, and empty fields.
Midlothian, Texas
Open Land → Google AI Data Center
South of Dallas, Google built a major data center campus on previously undeveloped land — part of the company’s multi-billion dollar expansion across Texas.
Abilene, Texas
Open Farmland → Stargate AI Campus ($500B Project)
Open farmland on the outskirts of Abilene, announced as the site of the largest AI infrastructure project in history.
New Albany, Ohio
Farmland → Google Data Center Campus
Ohio farmland northeast of Columbus transformed into one of Google’s expanding data center campuses — part of a wave that has made New Albany a major hub for cloud infrastructure.
DeKalb, Illinois
Farmland → Meta AI Data Center (2.3M sq ft)
505 acres of Illinois farmland became one of Meta’s largest data centers — a 2.3 million square foot facility completed in 2024, with a 560-acre expansion already proposed next door.
Power Consumption
Estimated electrical draw by featured site, in megawatts
Sources: Epoch AI facility database, operator filings, EIA interconnection records. 1 MW \u2248 800 homes.
How much power is that?
Megawatt numbers are easy to gloss over. Here's how the largest AI data centers compare to familiar references.
MW = megawatts of continuous power draw. A single large AI training campus now rivals an entire mid-sized American city — or a nuclear reactor — in electricity demand.
Where the Power Comes From
Hyperscalers publicize clean-energy commitments. But the grid each site actually pulls from tells a different story — dominated by natural gas in Texas and Ohio, nuclear in Illinois.
State-level electricity generation mix (EIA, 2024). Site-specific power purchase agreements may differ — hyperscalers often buy renewable credits to offset grid consumption, but the physical electrons still come from the local mix.
Built in a Hurry
Gigawatt-scale AI data centers are being built in months, not years — an order of magnitude faster than the infrastructure we historically built to power them.
xAI Colossus (in red) went from shuttered appliance factory to 100,000-GPU supercomputer in 122 days. A nuclear reactor of comparable output takes a decade or more. The grid wasn't designed for this pace.
Land Value Transformation
Assessed property value before acquisition vs. current estimated value
Sources: County assessor records, SEC filings, public land records. Values are estimated based on available data.
Who Pays
Behind every AI data center is a package of public subsidies — property tax abatements, infrastructure grants, payment-in-lieu-of-tax deals — negotiated with state and local governments eager for the headline investment figures.
Figures in USD millions, based on publicly reported agreements and available at time of announcement. Actual values vary with assessed property values, employment thresholds, and inflation adjustments. Sources: state EDC filings, Good Jobs First Subsidy Tracker, local news reporting. Values are approximate and should be verified from primary sources before publication.
The National Footprint
AI data center facilities across the United States — shape by operator, color by power draw
Interactive map
Set window.__MAPBOX_TOKEN or add a <meta name="mapbox-token"> tag to activate.
Size scales with power draw. Sites featured in the story above have a white halo and label.