Where AI Physically Lives — Data Center Geography and Power

1/ "The cloud" is a marketing term. AI physically lives in data centers — massive, power-hungry buildings in specific places chosen for specific reasons. Where AI lives determines who bears its costs. 🧵

2/ Northern Virginia hosts the largest data center cluster on Earth. "Data Center Alley" in Loudoun County processes an estimated 70% of the world's internet traffic. One county in Virginia is the physical backbone of the digital economy.

3/ Why Virginia? Fiber optic cables laid in the 1990s. Cheap land. Tax incentives. Proximity to Washington, DC. Infrastructure begets infrastructure. The geography of AI is path-dependent, not inevitable.

4/ The energy math is staggering. US data centers consumed an estimated 4-5% of national electricity in 2024, projected to reach 8-10% by 2028. That's equivalent to adding another New York City to the grid.

5/ In The Dalles, Oregon (pop. ~16,000), Google's data centers used over a quarter of the city's water in 2022. Residents had no say in the deal that brought the facility. The water was allocated before they knew what was happening.

6/ Ireland has become a European AI hub. By 2024, data centers consumed around 20% of the country's total electricity — more than all rural homes combined. The Irish grid operator warned of potential blackouts.

7/ The pattern: data centers locate where power is cheap and regulations are friendly. They bring tax revenue and some jobs (a $1B facility might employ 50 people). But they consume resources at a scale that reshapes local infrastructure.

8/ Kate Crawford mapped this in Atlas of AI: the "cloud" is lithium from South America, cobalt from the DRC, assembled in Asia, powered by coal in Virginia, cooled by river water in Oregon. It's a global extractive chain.

9/ The new frontier: AI companies are buying or building nuclear power. Microsoft signed a deal to restart Three Mile Island. Amazon bought a nuclear-powered data center campus. When you need gigawatts, you need dedicated generation.

10/ Geography creates winners and losers. Towns near data centers get tax revenue but strained grids. Remote workers in Nairobi label training data for cents. Users in San Francisco enjoy the AI product. The value chain is global; the costs are local.

11/ Some places are pushing back. Dutch provinces have imposed data center moratoriums. Singapore paused new builds for three years. Local resistance is growing as the physical footprint becomes impossible to ignore.

12/ Next time you use an AI tool, remember: your query travels to a specific building, in a specific town, drawing specific megawatts from a specific grid. "The cloud" has an address. The people who live there bear costs you never see. /end

---

LinkedIn version:

"The cloud" is a marketing term. AI physically lives in data centers — massive, power-hungry buildings in specific places, chosen for specific reasons. Where AI lives determines who bears its costs.

Northern Virginia's "Data Center Alley" processes an estimated 70% of global internet traffic. US data centers consumed 4-5% of national electricity in 2024, projected to reach 8-10% by 2028 — equivalent to adding another New York City to the grid. In The Dalles, Oregon (population ~16,000), Google's facilities used over a quarter of the city's water supply. In Ireland, data centers consume around 20% of national electricity, more than all rural homes combined.

The pattern is consistent: data centers locate where power is cheap and regulations friendly. They bring some tax revenue and few jobs (a $1B facility might employ 50 people), while consuming resources at a scale that reshapes local infrastructure.

Kate Crawford mapped this in Atlas of AI: the physical supply chain runs from lithium in South America to cobalt in the DRC, assembly in Asia, power from coal in Virginia, cooling from river water in Oregon. The value chain is global; the costs are local.

Some places are pushing back. Dutch provinces imposed moratoriums. Singapore paused new builds. The resistance grows as physical footprints become impossible to ignore.

Your AI query travels to a specific building, in a specific town, drawing specific megawatts from a specific grid. The people who live near that building bear costs most users never see. That's a redistribution problem worth understanding.