Artificial intelligence is widely represented as an immaterial, borderless, and essentially neutral technology. This paper challenges that representation by developing a framework of geopolitical externalization: the systematic displacement of AI infrastructure's real costs — environmental, social, and economic — onto territories and populations with the least governance capacity to resist them. Drawing on the proceedings of the interdisciplinary public conference Dalla Terra al Cloud (Trento, 13 May 2026), the paper traces the full supply chain of AI: from cobalt and coltan extraction in the Democratic Republic of Congo, through thirty years of resource-linked armed conflict and bilateral mineral agreements, to the energy, water, and labour demands of hyperscale data centre infrastructure. The paper further examines how the pay-per-token business model of AI firms creates perverse incentives that systematically increase resource consumption, and why efficiency gains cannot substitute for structural sufficiency policies. Against this material backdrop, existing international governance frameworks — including the EU AI Act — are shown to address AI primarily as a software problem, leaving its material infrastructure and the geopolitical inequalities it reproduces systematically unaddressed. The paper concludes by proposing situatedness — articulated through Federica Russo's ReDiEM framework of Relationality, Distributivity, Embodiment, and Materiality — as the governing principle for a more adequate international regulatory architecture.
Rafael S. Oliveira (Mon,) studied this question.