SoftBank put an unusual plan on the table on May 11: on the former Sharp site in Sakai, Osaka, it wants to build both an AI data center and its own battery production line.
Those sound like separate businesses. The logic appears once the site and the power problem are put together.
The land used to house Sharp's LCD panel factory. After Sharp wound down that business, the plant was left idle. SoftBank has bought the site and plans to split it into two linked facilities: an AX Factory for AI data center operations and AI hardware manufacturing, and a GX Factory for next-generation batteries and solar panels.
One site, two production systems
The AX side has a clear job: run AI agents. SoftBank's joint development program with OpenAI needs local compute, and Sakai is the first Japanese site in that plan. SoftBank has already talked about a much larger project, the 10 GW Stargate data center campus on U.S. Department of Energy land in Ohio. Sakai is smaller, but it is the first domestic version.
The GX side is the counterintuitive part. Why would a telecom and data center operator start making batteries itself?
The numbers explain the ambition. GX Factory is targeting annual capacity of 1 GWh, which would put it among the largest operations of its kind in Japan. Its commercial BESS units are planned at 140 to 560 kWh, while grid-scale systems range from 2,240 to 5,380 kWh. One technology partner, DeltaX, says it can fit 5 MWh of capacity into a 20-foot container.
SoftBank has not disclosed the total investment, but it has set a 2030 target of 100 billion yen in annual revenue, roughly 600 million dollars.
Mass production is scheduled for fiscal 2027, meaning after April 2027.
Why an AI data center wants its own batteries
The logic is simple: the scarcest resource for AI data centers is increasingly not chips, but electricity.
GPUs can be bought and racks can be stacked, but the grid may not keep up. Japan's grid was not designed over the past decade for a sudden wave of large data center loads. If SoftBank wants Stargate-class compute in Sakai, waiting for grid upgrades could become the bottleneck.
Battery energy storage systems sit in the middle. They can smooth peaks, provide backup power and let a data center discharge stored power during demand spikes, then recharge during low-demand periods. That reduces the instantaneous strain on the grid.
Tesla is pursuing a similar idea with Megapack, and LG Energy Solution is active in ESS in South Korea. SoftBank's decision to make batteries itself signals that it sees the production line as a strategic asset, not just a procurement item.
The partner list matters too. Cosmos Lab is working on zinc-halogen batteries with an aqueous electrolyte and claims the chemistry is less prone to fire. For data centers near urban areas, that matters: lithium battery fires have already pushed planners to tighten approvals.
A piece of the wider Stargate bet
SoftBank is launching the two factories together because its AI infrastructure wager is becoming more vertically integrated.
- Before the end of 2025, SoftBank plans to invest 22.5 billion dollars in OpenAI.
- The Stargate campus in Ohio is planned at 10 GW.
- The Sakai AX Factory is meant to run agents developed with OpenAI.
Put together, the chain is clear: invest in OpenAI, buy land in the United States and Japan, make batteries in-house and operate the data centers directly. It is a vertical integration strategy that reaches from the application layer down to energy infrastructure.
Masayoshi Son is not making a single-point bet. OpenAI's models have to win, Stargate has to come online on schedule, and the Sakai AX/GX architecture has to work. If any one part fails, the whole system gets weaker. If every piece works, the moat could be deeper than that of a standalone AI company.
Putting battery production and the data center on the same site is also unusual. A factory that runs AI while making its own batteries is not yet a common model anywhere in the world.
Details worth watching
- DeltaX's 5 MWh container design means the AX Factory's storage can be deployed in modules and expanded as demand grows.
- Annual capacity of 1 GWh sounds modest, but a mid-sized AI data center may use tens of MWh per day; that output could support storage for dozens of facilities.
- If Cosmos Lab can scale zinc-halogen technology, SoftBank could reduce dependence on lithium battery supply chains, including export controls on some key lithium-ion materials from China.
The next test is execution: whether production can ramp after 2027, and whether Sakai can prove the AX Factory model before Stargate formally comes online.
Sources: CocoLoop, SoftBank bets on battery building to back bit barns (The Register), SoftBank Plans to Make Large-Scale Batteries for AI Data Centers (Bloomberg), SoftBank launches Japanese battery storage business, production plant to house AI data center (Data Center Dynamics)