AI Data Centers Are Becoming a Critical-Minerals Demand Engine, New Study Finds

Highlights

• Copper accounts for 82–83% of total modeled mineral mass, with grid transmission driving 64% of that demand
• Grain-oriented electrical steel for transformers is identified as a major potential supply constraint
• For gallium, germanium, graphite, and rare earths, processing capacity and supply-chain concentration pose greater risks than geological scarcity
• Researchers built a bottom-up demand model covering 20 minerals across compute, thermal, and power systems from 2025 to 2035

A new study (opens in a new tab) finds that the rapid buildout of artificial intelligence data centers could become a significant new source of competition for critical minerals—not primarily because of semiconductors, but because of the massive power infrastructure required to support AI computing.

Lead author Macdonald Amoah, working with Maxwell Brown, Adam Simon, Morgan Bazilian, and Jahara Matisek, developed a bottom-up material-demand model covering 20 minerals from 2025 through 2035. The researchers separated AI infrastructure into compute, thermal-management, and power systems while distinguishing training, inference, and legacy data-center fleets.

The findings are striking: copper represents approximately 82%–83% of total modeled mineral mass, with grid transmission and distribution responsible for 64% of copper demand. Grain-oriented electrical steel (GOES), essential for transformers, also emerges as a major potential constraint.

For strategically important lower-volume materials—including gallium, germanium, graphite, lithium, cobalt, and rare earths—the researchers identify processing capacity and supply-chain concentration, rather than geological availability alone, as the greater vulnerability.

REEx Insight: AI is becoming another major competitor for already-constrained mineral and processing capacity. The study reinforces a central Rare Earth Exchanges® thesis: the strategic bottleneck increasingly lies downstream—in processing and refined materials—not simply in mining more ore.

Citation: Amoah, M., Brown, M., Simon, A., Bazilian, M., & Matisek, J. (2026). Mineral demand from AI data centers: Infrastructure intensity, processing bottlenecks, and supply competition. Resources Policy. DOI: 10.1016/j.resourpol.2026.105970.