AI data center capex supercycle reshapes infrastructure spending

The race to build AI-capable data centers is accelerating, with major investors and technology companies committing tens of billions to infrastructure expansion. SoftBank has invested billions in AI data center battery systems, while Blackstone and Halliburton are jointly investing $1 billion into VoltaGrid, a startup valued at over $10 billion. Circle Internet Group announced an AI infrastructure toolkit to power the agentic economy, signaling that infrastructure providers are positioning for sustained demand beyond traditional cloud capex cycles.

The buildout extends across the entire supply chain. Power distribution and cooling systems are becoming bottlenecks; energy companies like BE, FCEL, EOSE and AAON are rallying on expectations that AI data centers will require specialized cooling and power infrastructure. Optical networking companies are also benefiting, as high-capacity cabling and fiber are essential for connecting distributed AI clusters. Broadridge is deploying agentic AI at scale for capital markets, Flex CEO is spinning off a $6.5 billion AI infrastructure business, and utilities like Duke Energy are applying for Department of Energy loans to support grid modernization.

Copper is another critical input; every megawatt of hyperscale data center infrastructure requires roughly 27 tonnes of copper for transformers, substations and distribution systems. Copper prices have pushed to fresh 3-month highs near $13,619 per tonne, and mining companies (NEM, AEM, Codelco) are seeing renewed interest. The infrastructure buildout is creating a multi-year capex cycle that extends well beyond chip and GPU demand.

Risks include regulatory pushback on power consumption (some states are implementing environmental scrutiny), potential oversupply of capacity if enterprises slow deployment, and execution risk on the massive projects underway. Additionally, if AI economic returns disappoint or adoption stalls, capex could be curtailed sharply.