SoftBank Corp., the Japanese telecommunications arm of the broader SoftBank Group, announced on May 11 that it is launching a large-scale battery cell manufacturing business at its Sakai, Osaka facility. The plant will produce zinc-halogen battery cells in partnership with South Korea's Cosmos Lab and DeltaX. Target capacity is one gigawatt-hour of energy storage per year, scheduled to begin production in March 2028 with mass production targeted for 2029. The output is destined directly for SoftBank's own AI data center buildout.
The vertical integration is the point. SoftBank is not buying batteries on the merchant market. It is manufacturing them. The implication is that SoftBank views battery supply as a binding constraint on its AI data center pipeline serious enough to justify building captive manufacturing rather than depending on the global supply chain. That signal matters because SoftBank has been one of the most aggressive AI infrastructure investors of the last 18 months.
Zinc-halogen battery chemistry runs cooler than lithium-ion at comparable energy density, which is one of the technical attractions. But "cooler" is a relative term. A gigawatt-hour-scale battery installation co-located with AI data center loads still generates substantial waste heat during charge and discharge cycles, especially during the rapid load-following operations that a data center uses BESS for in the first place. Most existing battery installations rely on closed-loop chilled water or air cooling at the cell level.
For data centers, the cooling profile of co-located battery storage is now a non-trivial line item. AI data centers are already running into thermal management constraints at the rack level. Adding GWh-scale battery storage to the same site, drawing chilled water from the same plant, means the cooling capacity allocation between IT load and storage load becomes a design parameter. Cooling vendors selling into AI campuses now have to account for whether the customer has co-located battery storage that needs to share the thermal envelope.
SoftBank's decision is also a market signal. The company is one of the largest single customers in the AI infrastructure space and has direct exposure to grid limitations through its Japan operations. Japanese grid capacity for industrial loads is more constrained than U.S. grid capacity. SoftBank cannot solve its AI capacity problem by interconnecting more grid capacity, because that capacity does not exist. Battery storage that lets the company time-shift consumption against renewable generation is part of how the company makes the math work.
That model has implications outside Japan. Operators in markets where new substations and transmission lines have multi-year backlogs are increasingly looking at co-located BESS as the substitute for grid expansion. AWS, Microsoft, and Google have all expanded their behind-the-meter storage commitments through 2026. The cooling implications scale with battery scale: more storage on site means more thermal load to manage, more failure modes for the cooling plant to address, and more requirement for redundancy across battery and IT thermal envelopes.
The cooling vendor base has not yet fully absorbed the battery cooling opportunity. Most chiller plant designs treat BESS as an afterthought, allocating cooling capacity at the design phase but not optimizing the integration. SoftBank's Sakai investment indicates that GWh-scale battery storage is going to become a standard feature of AI data center campuses over the next five years. Vendors that can offer integrated cooling systems serving IT load, battery storage, and ancillary equipment from a single thermal management package have a competitive case to make.
The zinc-halogen chemistry specifically is worth tracking. If SoftBank's Sakai plant delivers a commercially competitive zinc-halogen product at gigawatt-hour scale by 2029, the broader battery storage market for data centers may pivot away from lithium-ion. Cooling architectures optimized for lithium-ion (lower thermal stability, tighter temperature windows) would not be the right match. Vendors should be modeling product roadmaps that can serve both chemistries during a transition that has now started but will not complete for several years.