AI Infrastructure Faces Copper and Power Constraints

While semiconductor supply has been the primary focus of AI infrastructure build-out, a secondary constraint is now surfacing: copper and cooling. Every megawatt of hyperscale data center capacity requires approximately 27 tonnes of copper for transformers, substations, power distribution, cooling systems, grid expansion and high-capacity cabling. As AI capex accelerates, copper demand from data centers is spiking hard.

Copper markets are already showing stress. Prices have pushed to a fresh 3-month high of $13,619 per tonne on the London Metal Exchange, only 6% below the January all-time peak near $14,500. Mining narratives are shifting accordingly. China's Zhaojin Mining is actively scouting for gold and copper acquisitions in Africa and Central Asia to scale production. The confluence of geopolitical risk (Iran war disrupting fertilizer and commodity flows) and structural AI demand is reshaping mining strategy globally.

Energy storage and cooling are equally critical. SoftBank has invested billions in AI data center batteries and power systems. Companies focused on energy storage (like Eos Energy Enterprises), fuel cells (Fuel Cell Energy), and cooling (Aaon) are seeing renewed investor interest. Duke Energy applied for Department of Energy loans representing potentially billions in customer savings for grid modernization, a sign that utilities are racing to expand capacity. The bottleneck is shifting from chips to the physical infrastructure that powers them: copper, electricity, and cooling capacity.

This creates opportunity and risk. Mining and energy infrastructure equities could see sustained demand. But it also signals that AI capex is hitting real-world constraints: if copper and power become the binding constraints, data center expansion slows and AI hyperscaler capex moderates. The narrative that 'AI will fix everything' is bumping up against physics and commodity scarcity.