Data center batteries and cooling emerge as next AI capex bottleneck

The semiconductor supercycle has obscured a second-order narrative: data center power and cooling infrastructure is becoming the next binding constraint on AI deployment. SoftBank recently invested billions in AI data center battery and energy storage companies, signaling to the market that Watt and BTU limitations are real and material. The thesis holds that hyperscalers cannot simply plug in more chips; they must simultaneously upgrade electrical grids, backup power systems, and thermal management. Stocks like Bloom Energy (BE), Fuel Cell Energy (FCEL), and Eaton (EATON) are positioned to benefit as data center operators scramble to secure reliable, low-latency power supplies.

Cooling is equally critical. Data centers running dense GPU clusters generate tremendous heat, and liquid cooling is becoming mandatory for next-generation hardware. Companies like Aaon (AAON) and smaller cooling integrators are fielding surging order books. The narrative gained traction after Nvidia and other chip firms announced optical designs and modular cooling architectures as part of their roadmaps. One energy analyst noted that the cost of cooling a modern AI data center can rival the cost of the silicon itself, turning thermal management into a boardroom issue.

This theme intersects with US energy policy and Trump's push for deregulation. Data center operators are lobbying for faster permitting of new power plants and substations. Energy importers like India and Turkey, already hit by the Iran war, face additional margin pressure as competing demand for electricity from AI data centers bids up wholesale power prices. Conversely, US energy producers and infrastructure firms stand to benefit. Defense contractors are also positioning themselves as "critical infrastructure" providers for data center security and redundancy.

The risk is that capex on batteries and cooling diverts dollars from semiconductor purchases, slowing the memory supercycle narrative. Alternatively, if power and cooling investments lag GPU deployments, hyperscalers could face utilization constraints, dampening the case for continued AI capex acceleration. For now, the market is pricing in abundant capital to address all these bottlenecks simultaneously, a view that may be too rosy if credit conditions tighten or if recession fears resurface.