AI's Hidden Mineral Crisis: The Next Bottleneck Isn't Chips-It's the Industrial Supply Chain

Highlights

• Copper accounts for 82–83% of total projected mineral demand from AI data centers, with grid infrastructure consuming nearly two-thirds of that total.
• Grain-oriented electrical steel and rare earth permanent magnets emerge as critical but overlooked bottlenecks in AI infrastructure expansion.
• The study concludes that processing and refining capacity—not geological scarcity—poses the greatest supply risk for strategic minerals.
• Rare earth NdFeB magnets enable high-efficiency cooling and power systems in AI data centers, driving demand for magnet manufacturing capacity.
• Investors are advised to look beyond semiconductor companies toward processing, metallurgy, grid infrastructure, and integrated mine-to-manufacturing ecosystems.

A study (opens in a new tab) published in Resources Policy, and led by Macdonald Amoah, an independent researcher with co-authors Maxwell Brown, Morgan Bazilian, and Jahara Matisek with Colorado School of Mines as well as Adam Simon, University of Michigan, concludes that artificial intelligence data centers will reshape demand for critical minerals (and rare earth elements)—but not primarily through AI chips. Instead, power infrastructure, electrical equipment, and processing capacity become the dominant constraints. Key findings: Copper dominates projected mineral demand, while grain-oriented electrical steel (GOES) emerges as another critical bottleneck. The authors conclude that, for several strategic minerals—plus rare earths—the greatest supply risk lies in processing, not geological scarcity. Rare Earth Exchanges® view: This paper independently reinforces a core REEx thesis: in Great Powers Era 2.0™, industrial capability—not ore—is becoming the decisive strategic asset. Although, of course, lots of quality feedstock will be necessary.

Did You Know

Rare earth permanent magnets—primarily neodymium-iron-boron (NdFeB) magnets—play a critical but often overlooked role in modern AI data centers. Rather than powering computer chips directly, they enable the high-efficiency electric motors used in cooling fans, liquid-cooling pumps, HVAC systems, and certain precision data storage and transmission equipment. Their exceptionally strong magnetic properties allow motors to deliver greater torque and airflow while remaining smaller, lighter, and more energy efficient—an increasingly important advantage as AI servers become more power-dense and generate substantially more heat. As hyperscale AI infrastructure expands, demand for these high-performance magnets is expected to rise alongside the need for reliable cooling and power systems, further increasing competition for rare earth processing and magnet manufacturing capacity.

AI's Hidden Appetite Isn't Chips—It's the Industrial Supply Chain

Everyone wants to talk about Nvidia. The paper’s authors argue investors should spend more time thinking about transformers, substations, copper, and processing plants. In a new Resources Policy study, (opens in a new tab) lead author Macdonald Amoah and colleagues develop a bottom-up engineering model that projects mineral demand across 20 critical materials used in AI data center infrastructure through 2035. Their conclusion is striking: AI is fundamentally an infrastructure buildout, not merely a semiconductor story. The greatest long-term supply risks arise less from mining than from the industries that transform raw materials into usable products.

The Story Beneath the Servers

The model finds that copper accounts for roughly 82–83% of total modeled mineral demand, with grid transmission and distribution consuming nearly two-thirds of that total. Grain-oriented electrical steel also emerges as a significant enabling material. For lower-volume strategic minerals—including gallium, germanium, graphite, lithium, cobalt, and rare earths—the authors identify downstream processing capacity as the principal vulnerability rather than raw resource availability.

That distinction matters.

Macdonald Amoah, First Author

The paper does not argue that AI will create a global rare earth shortage. Rather, it concludes that AI introduces another major source of competition for already constrained refining, metallization, and advanced manufacturing capacity.

Why Investors Should Pay Attention

Rare Earth Exchanges has consistently argued that the strategic race is shifting from mines to industrialization. This study provides independent academic support for that view. It also highlights by-product recovery from existing U.S. mining operations as one practical avenue for expanding domestic mineral supply without waiting for entirely new mines to enter production.

For investors, the lesson is straightforward: AI strengthens the investment case not simply for mining companies, but for processing, metallurgy, specialty materials, grid infrastructure, and integrated mine-to-manufacturing ecosystems. In Great Powers Era 2.0, value increasingly accrues to those who can transform minerals—not merely extract them.

Amoah M, Brown M, Simon A, Bazilian M, Matisek J. Mineral Demand from AI Data Centers: Infrastructure Intensity, Processing Bottlenecks, and Supply Competition. Resources Policy. 2026. doi:10.1016/j.resourpol.2026.105970.