AI infrastructure requires massive quantities of metals, particularly copper and silver (volume bottlenecks) and gallium/rare earths (chokepoint bottlenecks), with AI data centers alone consuming 400,000+ tonnes of copper annually by 2030s. The global copper market faces significant supply deficits (124,000-304,000 tonnes in 2025-2026, potentially 6-10 million tonnes by 2040), making metals the real constraint for AI scaling rather than just GPUs or power.