Metals for AI Hardware and Cooling: A Multi-Metal Demand Model, 2026-2030

Demand from AI data center construction is becoming a structurally distinct driver for at least six metals at once, separating AI hardware exposure from broader electrification and consumer-electronics cycles. This report models incremental annual tonnage by metal and application layer: copper in power distribution, busbars, and liquid-cooling cold plates; aluminum in chassis and heat exchangers; gold and silver in GPU bonding wire, substrate interconnects, and thermal interface materials; and neodymium-praseodymium plus dysprosium in cooling-system magnets. The supply constraints are asymmetric, with copper facing a sizeable refined deficit and rare earth exposure concentrated in Chinese processing. Projections are anchored to disclosed hyperscaler capex, GPU shipment forecasts, and the adoption curve for direct-to-chip liquid cooling.