SoftBank Pivots To Power AI Data Centers

SoftBank Group is pivoting directly into the power side of AI, launching a new battery and energy storage business explicitly aimed at data centers. The initiative targets gigawatt‑hour scale battery cell and storage production around FY2028 and an annual revenue run‑rate of roughly ¥100 billion from its domestic battery operations.

With AI workloads driving sharply higher rack densities and electricity consumption, power availability and thermal management are emerging as binding constraints for data center expansion. Historically, when a specific infrastructure layer becomes the bottleneck, capital and profit pools have shifted toward that layer, as seen in network equipment during the late‑1990s Internet build‑out and carrier‑neutral data centers in the early cloud era.

Under similar conditions, specialized power and cooling suppliers have often outperformed, exemplified by Equinix during the public cloud ramp and Qualcomm in mobile broadband as RF and baseband became rate‑limiting. The current focus on AI‑driven power constraints places batteries, energy storage, and advanced cooling in a comparable position within the AI stack.

Equipment and infrastructure providers directly leveraged to these constraints include Vertiv Holdings (VRT), Eaton (ETN), and Schneider Electric (SUp), which supply uninterruptible power systems, electrical distribution, and liquid cooling architectures for dense AI deployments. Data center landlords such as Equinix (EQIX) monetize the same bottlenecks via scarce high‑power, high‑cooling capacity, with pricing power strongest in markets where grid connections and thermal capacity are hardest to add.

The pattern remains conditional: past cycles have shown that once large pools of capital crowd into a bottleneck, subsequent over‑build can quickly compress returns. SoftBank’s move nonetheless reinforces the market signal that AI data center power, storage, and cooling are shifting from background utilities to front‑of‑stack constraints that can reorient where infrastructure economics accrue within the AI ecosystem.