Artificial intelligence companies are driving a new era of imperial extraction by utilizing low-wage labor in the Global South and depleting critical natural resources in developing nations to power Large Language Models (LLMs), according to reporting by technology reporter Karen Hao in her work Empire of AI. This system relies on a “ghost workforce” for data labeling and the massive consumption of freshwater for data center cooling, often at the expense of local communities in regions like Kenya and Chile.
The scale of this operation is not merely a byproduct of technical growth but a structural necessity. To make AI appear “intelligent,” companies like OpenAI require millions of hours of human intervention to filter toxicity, correct errors, and categorize data. This process, known as Reinforcement Learning from Human Feedback (RLHF), is frequently outsourced to firms in East Africa and Southeast Asia where labor laws are lax and wages are minimal.
How the “Ghost Work” Economy Operates in Kenya
In Kenya, the AI industry has established a pipeline of precarious labor. Workers are hired via third-party contractors to review traumatic content—including images of violence and abuse—to train safety filters for models like GPT-4. These workers often earn pennies per task, lacking the psychological support or benefits provided to the engineers in San Francisco.
This labor model mirrors the historical “piece-work” systems of the industrial revolution. By distancing themselves from the workers through layers of subcontractors, AI giants avoid direct employer liability. According to TIME Magazine, this creates a systemic vulnerability where workers can be deactivated from platforms without notice or recourse, effectively treating human intelligence as a disposable raw material.
The economic disparity is stark. While OpenAI’s valuation has soared into the hundreds of billions, the individuals refining the very data that drives that value remain in a cycle of digital poverty. This is not a “partnership” with the Global South, but a unidirectional extraction of cognitive labor.
Why Data Centers are Draining Chilean Water Tables
The extraction extends beyond labor to the physical environment. AI models require immense computational power, which generates heat that must be managed through cooling systems. In Chile, the push for data center infrastructure has led to attempts to secure massive quantities of freshwater from communities already struggling with drought and water scarcity.
Water is the invisible fuel of the AI boom. When these facilities are placed in water-stressed regions, the "cost" of a prompt is not just electricity, but the potential dehydration of local agriculture.
This creates a geopolitical tension where the “intelligence” of the Global North is subsidized by the ecological degradation of the Global South. The Chilean government has faced increasing pressure to regulate how tech firms access aquifers, as the thirst of the cloud threatens the survival of rural townships.
The Geopolitical Blueprint of AI Colonialism
This phenomenon is often termed “AI Colonialism.” It is the application of colonial-era extraction patterns—taking raw materials (data and minerals) and cheap labor from the periphery to create high-value products for the center—updated for the silicon age. The “raw materials” here are not just cobalt or lithium, but the very linguistic and cultural nuances of human speech.
The winners in this scenario are the hyperscalers—Google, Microsoft, and Amazon—who control the compute and the capital. The losers are the “data laborers” and the displaced residents of water-stressed zones. This creates a feedback loop: the more the AI evolves, the more “human” data and cooling it requires, intensifying the pressure on the Global South.
What Happens When the Resource Gap Collapses?
The industry is reaching a tipping point. As the “easy” data is exhausted, AI companies are hunting for more niche, high-quality human feedback, which will likely push extraction into even more marginalized communities. Simultaneously, the climate crisis is making the water-heavy cooling models unsustainable.
According to the United Nations Environment Programme (UNEP), the intersection of digital expansion and environmental collapse requires a fundamental shift in how “innovation” is measured. If a model’s efficiency increases but its water footprint doubles, the net global impact is negative.
The industry’s response has been largely performative, focusing on “carbon offsets” while ignoring the immediate, localized theft of water and labor. The transition to “green AI” remains a corporate talking point rather than a deployed reality in the regions where the damage is most acute.
The AI revolution is often marketed as a tide that lifts all boats. However, for the workers in Nairobi or the farmers in Chile, that tide looks more like a flood of extraction. We have to ask: is the convenience of a chatbot worth the systemic exploitation of the people who actually build its mind? If we don’t demand transparency in the AI supply chain, we aren’t witnessing progress—we’re witnessing a rebranding of empire.
Do you think the convenience of AI justifies the hidden human and environmental costs? Let us know in the comments.
