Artificial Intelligence at What Cost?

This paper inventories the full cost of artificial intelligence, from the minerals in a singlerack to the thermodynamics underneath the entire buildout. It begins with the supply chain as itstands — accelerator concentration, the three-company memory oligopoly, foundry andpackaging chokepoints, the ODM assembly layer, and the materials (copper, aluminum, cobalt,lithium, nickel, graphite, rare earths) that go into the machines — then follows the constraintsthat decide whether that chain can keep scaling: transistor physics, the AC-to-DC conversion tax,water treatment, the 2026–2028 balance between surplus raw tonnage and sold-out finishedinputs, the human and environmental cost of mining, climate and diesel, and heat reuse. Itcontinues through the software moat, the decommissioning wave, and the rest of AI: models,data, the people who label it, the financing, the three-bloc governance split, safety, history,recursive AI, and children whose critical-thinking steps are skipped when the work is offloaded. Itcloses with the principle underneath all of it: efficiency does not eliminate cost, it moves it — fromchip to cooling tower to river to atmosphere — and the bill, diffused and delayed, always arrives.All figures are drawn from public sources.1. Anatomy of a rack — w