SoftBank is betting that the real money in AI isn't just training models or running inference—it's keeping the lights on when the grid can't.

The Summary

The Signal

SoftBank Group's mobile unit is pivoting from telecom carrier to power supplier, manufacturing large-scale battery cells at a facility in Sakai, Osaka. The move acknowledges what everyone building AI infrastructure already knows: the constraint isn't compute anymore. It's watts.

AI data centers don't just need more power. They need reliable, instantaneous power that can handle the spiky, unpredictable loads of training runs and inference bursts. Grid power works for steady-state operations. But when a cluster spins up a new training job or scales inference capacity in milliseconds, you need batteries that can deliver massive current without blinking.

"SoftBank is building the infrastructure layer that makes AI infrastructure possible."

The company isn't stopping at batteries—it's also planning to sell AI computing power itself. That's the real tell. SoftBank sees the value chain clearly:

  • Make the batteries that stabilize the power
  • Sell the compute that runs on that power
  • Capture margin at both ends of the stack

This is vertical integration for the agent economy. The companies winning in Web4 won't just build software or own data centers. They'll own the electrons.

The Implication

Watch for the battery manufacturers who already supply EV markets to follow SoftBank into data center energy storage. The physics are similar, but the margins and contracts are better. Data centers sign long-term power agreements. Car buyers just want range.

If you're building AI infrastructure or running agent workloads at scale, battery-backed power is about to become table stakes. The question isn't whether you need it. It's whether you'll pay SoftBank's price or build your own.

Sources

Bloomberg Tech