Extractive AI

Executive Summary

Although artificial intelligence models vary greatly in concept and application, they all intensify extractivism and impose an environmental toll wherever they are harnessed or deployed. AI's extractivism goes beyond mining data from the internet and plundering the work of artists and writers: it is rooted in deep ties to the mining and fossil fuel industries and to the capitalist system that enables its extraction. Training and running AI require large amounts of data and computing power delivered through hyperscale data centres leased by firms like Amazon, Microsoft, and Google. These infrastructures currently consume somewhere between 1-4% of global energy supplies, a figure driven upward year after year by cryptocurrencies and AI, while consolidating control of critical computing infrastructure among a handful of tech and investment giants.

The essay argues that AI's harms cannot be resolved through efficiency accounting, emissions reporting, or "frugal AI" optimization. Training GPT-3 alone required roughly 700,000 liters of clean freshwater to keep servers from overheating, and scaling AI drains lakes and rivers in water-stressed regions, exacerbating droughts, displacing vulnerable populations, and diverting public water toward private tech profits. Minerals essential to AI, such as neodymium, cerium, praseodymium, cobalt, nickel, and lithium, are frequently mined in conflict zones with lax labour regulations using methods that contaminate groundwater and destroy habitats.

Ultimately, the authors contend that AI functions as an agenda-setting industry rather than a discrete technology, serving as a marketing tactic for the billionaire and trillionaire class. Policymakers and the public should be skeptical of appeals to "artificial general intelligence" and existential-risk narratives, which distract from AI's present environmental impact and its entanglement with the fossil fuel, mining, real estate, energy, and military industries. Meaningful policy must tackle the underlying social and economic factors, not merely make AI marginally more "sustainable."