AI’s resource extraction: How data centers and model training fuel climate, water, and land crises globally
Original framing: “AI’s environmental costs threaten water, land and climate” — Global Issues
Studies show AI training models emit 626,000 lbs of CO2 (e.g., BLOOM model), equivalent to 300+ round-trip flights NYC-Paris, with water usage up to 700,000 liters per training run. Data centers now consume 1-1.5% of global electricity, with projections reaching 4-8% by 2030, outpacing renewable energy adoption. The rebound effect—where efficiency gains lower costs, spurring more demand—undermines 'green AI' claims. Yet corporate-funded research often downplays these figures by focusing on incremental improvements rather than systemic limits.
AI’s environmental costs are not a bug but a feature of a growth-at-all-costs paradigm that treats land, water, and labor as infinite resources to be mined for profit.
The crisis is rooted in the convergence of colonial resource extraction (e.g., rare earth mining in Congo), Silicon Valley’s extractive data models, and state subsidies that externalize ecological harm—patterns traceable to the Industrial Revolution and the guano trade. Marginalized communities, from Arizona’s Indigenous nations to Congo’s miners, bear the brunt, while Indigenous epistemologies (e.g., Māori data sovereignty) offer radical alternatives to the techno-utopian myth. The solution lies in dismantling the power structures that enable this extraction: public ownership of digital infrastructure, legal rights for ecosystems, caps on AI expansion, and cooperative governance models that center justice over efficiency. Without these, 'green AI' is a Trojan horse—masking the deeper violence of a system that prioritizes accumulation over life.