SoftBank to Build Gigawatt-Scale Zinc-Halogen Battery Plant in Osaka for AI Data Centers

SoftBank announced on May 11 that it will manufacture zinc-halogen battery cells at its Sakai City site in Osaka through a partnership with Korea’s Cosmos Lab. Production is scheduled to begin in fiscal year 2027, ramp to gigawatt-hour-scale annual output by fiscal year 2028, and reach ¥100 billion in revenue by fiscal year 2030. The cells use a pure-water electrolyte and contain no lithium and no cobalt.

The site is the same Sakai City campus that will host a large-scale AI data center and a separate AI hardware plant. SoftBank is not building a battery factory and shopping for an offtaker. It is co-locating cell manufacturing with the load it intends to serve.

The chemistry pitch. Zinc-halogen flow and semi-solid systems have circulated on the periphery of the storage industry for years. What is new in the SoftBank announcement is that a balance-sheet-grade buyer of data center power is willing to underwrite gigawatt-hour-scale capacity for a non-lithium chemistry whose entire marketing premise is the elimination of thermal runaway risk. The Japan Times, Data Center Dynamics, and ESS News all framed the announcement around fire safety rather than energy density, round-trip efficiency, or cost.

That framing reflects where the procurement conversation has moved. NFPA 855’s 2026 edition, finalized last quarter, mandates dedicated battery rooms for indoor lithium-ion installations above 600 kilowatt-hours, eliminates the hazardous materials assessment exemption, and requires thermal runaway propagation prevention systems. UL 9540A’s testing burden has expanded across successive editions. The cost of an indoor lithium installation in a data center, hospital, hospitality property, or downtown office tower is no longer dominated by the cells. It is dominated by the room around the cells.

A chemistry that does not require the room changes the math.

The timeline matters more than the chemistry. Fiscal year 2028 in Japan ends March 2029. Gigawatt-hour scale by that date would put zinc-halogen production roughly three years behind the current generation of lithium iron phosphate gigafactories now coming online in Tennessee, Kentucky, and Nevada. Cosmos Lab has not yet published commercial deployment data at the scale SoftBank is committing to, and the Sakai City site has not begun construction.

The structural signal is not that zinc-halogen is about to enter the US commercial behind-the-meter market. It is that a hyperscale-adjacent buyer has decided the indoor-safety problem is worth solving with a new chemistry rather than with engineering controls layered on top of lithium. That decision sets the procurement frame for every chemistry-agnostic buyer evaluating a commercial battery purchase between now and 2028.

Where the announcement intersects US commercial storage. Commercial buyers in the United States are not waiting for zinc-halogen. They are buying lithium iron phosphate systems today and accepting whatever indoor compliance pathway their local authority having jurisdiction approves. The NFPA 855 2026 rules push that compliance cost upward; the UL 9540A indoor-rated systems already on the market, including the Viridi 480-volt cabinet listed under UL 9540 in April and SolarEdge’s 197-kilowatt-hour CSS-OD released in Europe the same week, define the certified competitive set.

SoftBank’s announcement reframes that competitive set as a category whose existence depends on lithium remaining the only commercial option. If a non-lithium chemistry can plausibly enter at gigawatt-hour scale by 2028, the indoor-certification moat that current incumbents have built around fire codes becomes a moat with a defined expiration date rather than an open-ended one. Procurement officers running multi-year capital plans for hospital systems, university campuses, and large hotel portfolios will begin to factor that into rate-of-return assumptions on lithium purchases made today.

That does not slow today’s deployments. Demand-charge math on commercial rate schedules still pays back faster than the zinc-halogen commercial timeline. But it does change how a long-duration procurement decision gets framed in front of a CFO.

The other read: domestic chemistry diversity is becoming a procurement requirement. A Utility Dive piece published the same day from Torus CEO Nate Walkingshaw argued that storage competition has shifted away from cell chemistry breakthroughs and toward integrated system capabilities, deployment speed, cybersecurity, and supply chain resilience. The SoftBank announcement complicates that thesis. Chemistry is becoming a procurement variable again, not because of energy density, but because of siting flexibility.

Indoor approval pathways, distance-to-exit requirements, and insurance underwriting all read chemistry. A pure-water electrolyte system writes itself into a permitting envelope that lithium iron phosphate cannot. Whether that envelope translates to lower installed cost depends on factory throughput, balance-of-system maturity, and round-trip efficiency, none of which Cosmos Lab has demonstrated at scale.

The Foreign Entity of Concern rules layered onto the Investment Tax Credit add a second variable. A Korean cell supplier manufacturing in Japan sits outside the FEOC perimeter the way Chinese cells do not. If zinc-halogen production at Sakai City is structured to qualify for any US import or licensing pathway by 2028, the chemistry’s regulatory profile is more favorable than its main lithium iron phosphate competitor on a domestic content adder basis.

The forward question. Indoor commercial storage has spent the past three years consolidating around UL 9540A as the de facto certification standard for lithium iron phosphate. The Viridi and SolarEdge listings show the market accepting that lithium can be made compliant if the integration work is done correctly. The SoftBank bet is that the integration work is the constraint, and that a chemistry with no thermal runaway pathway removes the constraint entirely.

Three years is enough time for that bet to be wrong, right, or technically correct but commercially irrelevant. CATL, BYD, and the new US lithium iron phosphate plants will not stand still during the same window. Battery pack prices fell to $108 per kilowatt-hour in 2025 even as lithium carbonate spot prices more than doubled, which means cell-margin compression has more room to run before incumbents need to defend their position with a chemistry change.

What the announcement does establish is that the indoor-safety conversation has graduated from a fire-code question to a chemistry-procurement question. That graduation will shape every commercial storage purchase order written in 2027 and 2028, regardless of which chemistry the buyer eventually picks.

Sources