AI capex stalls on copper and power crunch

The race to build AI infrastructure has exposed a brutal reality: the compute boom cannot outrun the real world. Hyperscalers need 27 tonnes of copper per megawatt of data center capacity for transformers, substations, power distribution, cooling systems, and grid expansion. Copper spot prices have surged to fresh three-month highs near USD 13,619 per tonne, only six percent below the January all-time peak of USD 14,500. Mining stocks like Teck Resources, Freeport-McMoRan, and Nevada Copper are seeing renewed interest as the market wakes up to scarcity.

Beyond copper, the energy constraint is becoming equally acute. Observers note that next-generation data centers need massive cooling and storage systems. Softbank has invested billions specifically in AI data center battery infrastructure. Small-cap energy and battery plays like Eos Energy Enterprises (EOSE), Fluence (FLNC), and fuel cell makers like FuelCell Energy (FCEL) are attracting institutional capital as solutions. Analyst Marc Champion highlighted the CV90 armored vehicle manufacturing shift as evidence that Europe is improving defense coordination; by parallel, logistics and manufacturing for energy infrastructure are becoming critical industrial assets.

The implications ripple across commodities and utilities. Energy importers face margin pressure from elevated oil and gas; defense names and infrastructure plays benefit from the elevated risk premium and visibility of capex cycles. Semiconductor manufacturers are also caught in the vise: higher input costs for power and copper eat into chip margins, even as AI chip demand remains robust. This dynamic could force a broader re-rating of the AI supercycle, from frothy software multiples toward heavy industrials and raw materials.

Skeptics argue that efficiency gains and alternative materials (e.g., optical or non-copper interconnects) will eventually relieve the bottleneck. However, near-term supply is inelastic, and mining production cannot scale overnight. The narrative is shifting from a pure supply-demand AI chip story to a more complex one about the physical infrastructure required to deploy AI at scale.