AI capex race accelerates with infrastructure deals

Behind the semiconductor rally lies a genuine structural shift in capex allocation toward AI compute and supporting infrastructure. Flex Ltd. CEO Revathi Advaithi is leaving her post to lead Flex's AI infrastructure spinoff valued at $6.5 billion, signaling the magnitude of this buildout. Blackstone and Halliburton jointly invested $1 billion into VoltaGrid, a power-infrastructure startup, recognizing that AI data centers face acute energy constraints. Circle Internet Group launched an AI infrastructure toolkit designed to power agentic economy applications. These moves reflect a real bottleneck: compute capacity and the power, cooling, and real estate to house it are the limiting factors in AI scaling, not silicon alone.

The infrastructure race is creating secondary beneficiaries beyond chip makers. Battery and energy-storage names like SoftBank-backed projects, fuel-cell developers, and cooling integrators are seeing institutional buying. Real estate investment trusts focused on data center real estate, energy transitioning plays, and materials-science firms supplying advanced cooling and power management are rotating higher. This is distinct from the chip rally; it addresses the physics of running AI at scale: energy density, cooling, land, and grid capacity all matter.

The narrative of infinite AI capex is now priced into equities, but supply-chain constraints in semiconductors, power infrastructure, and rare-earth materials could create bottlenecks. If cooling or power grid capacity fails to keep pace with demand, capex growth will stall and valuations could compress. Additionally, if AI monetization disappoints relative to capex investment, a capex recession could follow sharp earnings misses in 2026-2027.

This supercycle story is partially real but faces execution risk. Investors cheering the spinoffs and infrastructure deals are betting on perpetual capex growth, a wager that depends on AI productivity gains translating to massive revenue growth for hyperscalers. So far, ROI data on AI capex is sparse. The infrastructure narrative provides cover for the chip rally, but if data center utilization rates decline or major hyperscalers announce capex cuts, the entire AI complex unwinds.