THE ECOLOGICAL COST OF AI
Mapping the invisible infrastructure of the „Cloud“
Artificial Intelligence is not ethereal. It is built from rocks, powered by rivers, and cooled by the atmosphere. This dossier prepares a Google Earth expedition across the planetary supply chain of AI, identifying specific sites of extraction, production, operation, and resistance.
1. The Material Chain
Before visiting specific coordinates, we must understand the linear „Take-Make-Waste“ flow. Every ChatGPT query travels through this physical infrastructure.
The Cobalt Bottleneck
AI hardware requires massive battery backup and specific circuitry. The DR Congo supplies the vast majority of the world’s cobalt, often involving informal (artisanal) mining with high social costs.
The Thirst of Chips
A single cutting-edge semiconductor factory consumes millions of gallons of ultrapure water daily. This creates conflict in arid regions like Arizona and Taiwan.
Projected AI Energy Demand (TWh)
Computational power for training models is doubling rapidly. By 2026, data centers could consume as much electricity as Japan.
3. Global Site Selection
Use these coordinates to pilot your Google Earth presentation. Each card represents a critical node in the AI value chain.
Mutanda Mine
-10.7815, 25.8082
View: Satellite zoom on open pit & acid pools.
One of the world’s largest Cobalt mines. Look for green acid ponds vs surrounding deforestation.
Salar de Atacama
-23.5000, -68.3333
View: High altitude satellite.
Vibrant blue evaporation ponds for Lithium. Contrast with the extreme aridity of the surrounding desert.
Morowali Industrial Park
-2.8275, 122.1553
View: Coastline industrial sprawl.
Nickel processing center. Observe coal power plants powering the facility and tailings runoff.
TSMC Fab 18 (Tainan)
23.1135, 120.2764
View: 3D Building view of the massive campus.
Produces 3nm chips. Look for water treatment facilities required for the drought-prone region.
Intel Ocotillo Campus
33.2386, -111.8747
View: Satellite view of complex in desert.
Example of water-intensive industry in water-scarce environment. Look for cooling towers.
Data Center Alley (Ashburn)
39.0438, -77.4874
View: Street View/Flyover along Waxpool Rd.
Highest concentration of data centers globally. Endless windowless grey boxes replacing farmland.
Meta Luleå Data Center
65.6322, 22.0734
View: Satellite view near Arctic Circle.
Built for natural cooling. Highlights the geographic race for thermal management.
Dublin Hyperscale Cluster
53.3075, -6.4431
View: Grange Castle Business Park.
Microsoft/Google facilities causing national grid instability and energy debates.
Agbogbloshie (Old Site)
5.5500, -0.2260
View: Satellite history or nearby lagoon.
Formerly world’s largest e-waste dump. Look for scorched earth from cable burning.
Guiyu (Recycling Hub)
23.3275, 116.3556
View: Street view (historical) or dense industrial blocks.
Historically the E-waste capital. Shows transition from informal to semi-formal industrial processing.
Cerrillos Data Center Site
-33.5042, -70.7186
View: Urban area near aquifer.
Location of successful community protests against a Google data center due to water concerns.
Zeewolde (Meta Site)
52.3333, 5.5000
View: Agricultural Polder lands.
Site where the Dutch senate blocked a massive Hyperscale center to protect green energy limits.
Synthesis & Reflection
Key Theses
- Materiality: AI is not virtual; it is geological. It moves mountains in the Congo and drains aquifers in Chile.
- Asymmetry: The benefits of AI concentrate in the Global North, while the toxic externalities (mining, e-waste) are exported to the Global South.
- Energy Paradox: AI optimization claims to solve climate change, yet its own infrastructure currently accelerates energy consumption.
- Water Intensity: „Cloud“ computing is thirsty. Large models „drink“ freshwater during training and inference.
Discussion Questions
- How does visualizing the physical location of a „cloud“ server change your perception of using tools like ChatGPT?
- Is it ethical to deploy water-intensive AI infrastructure in regions facing drought?
- Who should bear the cost of the e-waste generated by the rapid obsolescence of AI hardware?