Key Takeaways
- 1 Electricity—not capital or chips—is the primary constraint on AI growth
- 2 Gigawatt-scale loads require transmission lines, not urban power grids
- 3 Grid interconnection takes 5-8 years—longer than building a data center
The Scale Problem
The U.S. generates 4,000+ TWh annually. Data centers use ~5%. So why the crisis? Because electricity isn't fungible—you can't move it freely or store it economically.
24/7 Maximum Demand
Grids are built for peak demand—the highest load that might occur. Traditional loads are "peaky." Data centers are constant.
Peaky Demand
Varies by time, season, weather
Constant Peak
90-95% utilization, 24/7
Why Not Cities?
Cities have power and fiber. But they have distribution infrastructure—designed for neighborhoods, not power plants.
12-35 kV
Homes, offices, light industrial
Max: ~100 MW
115-765 kV
Power plants, regional substations
Capacity: 1,000+ MW
Site selection follows transmission lines, not fiber networks.
Farmland near high-voltage corridors beats urban real estate.
The Interconnection Bottleneck
Connecting to the grid isn't just plugging in. It requires years of studies and upgrades.
Following the Wires
Understanding the power constraint explains the geography of mega-projects.
Not Silicon Valley. Not New York.
Transmission access and interconnection speed trump tech ecosystems.
Go Deeper
Chapter 4 of This Is Server Country traces how electricity became the limiting factor in AI infrastructure—exploring interconnection queues, transmission networks, and the policy challenges of building gigawatt-scale facilities in rural communities.
Learn more about the book →