SoftBank battery business links AI data centers to storage

Overview

SoftBank battery business plans are turning AI infrastructure into a storage-manufacturing question. SoftBank Corp. said on May 11 that it will build a Japan-based battery business around battery cells and energy storage systems, using its planned Osaka Sakai AI Data Center as the core hub for a wider industrial cluster.

The move matters because it treats batteries as part of the AI buildout itself. Data centers are already forcing utilities, regulators, and developers to rethink power supply. SoftBank's plan adds another layer: companies building AI infrastructure may start manufacturing storage, solar-related equipment, and energy systems to control more of the power chain.

SoftBank battery business starts at the Sakai site

SoftBank's official announcement is the anchor fact. In its May 11 battery business release, SoftBank Corp. said it launched a Japan-based battery business to support rising electricity demand driven by AI adoption. The company plans to use the former Sharp factory site in Sakai City, Osaka Prefecture, where it is developing the Osaka Sakai AI Data Center, as a core hub.

The site will host two related factory concepts. SoftBank describes the AX Factory as a hub for AI data-center operations and AI infrastructure hardware manufacturing. The GX Factory will focus on next-generation batteries, solar panels, and related products.

That pairing is the real development. SoftBank is not only buying batteries for an AI data center. It is planning a physical cluster where data-center operations, AI hardware, battery cells, storage systems, and solar-linked manufacturing sit in the same strategic frame.

AI data center batteries move from procurement to strategy

For years, large data-center operators mostly treated power as something to procure: utility deals, power-purchase agreements, backup generators, interconnection queues, and renewable-energy certificates. The AI data-center boom is changing that. Power is now a core constraint on where computing can be built and how fast it can scale.

SoftBank's plan reflects that shift. The company said the battery business will cover development through manufacturing and will produce battery cells and battery energy storage systems with advanced technologies. Production is planned to begin in the fiscal year ending March 31, 2028, with mass production around a gigawatt-hour per year targeted by fiscal 2028.

The Wall Street Journal reported that SoftBank is working with South Korea-based Cosmos Lab and DeltaX, with commercial production expected to start in fiscal 2028 and mass production following the next year. The WSJ report on SoftBank's battery venture also said the company aims for annual revenue above 100 billion yen from the battery business by fiscal 2030.

That revenue target makes the move bigger than internal backup planning. SoftBank is positioning storage as a business line tied to AI power demand.

Zinc-halogen batteries are the safety bet in the plan

The most distinctive technology claim involves zinc-halogen batteries. SoftBank said it is collaborating with COSMOS LAB to develop battery cells that combine non-flammable characteristics with strong storage performance. The company said the cells use pure water as the electrolyte, reducing ignition risk compared with dominant lithium-ion batteries that use flammable organic electrolytes.

SoftBank also said the chemistry uses halogen material for the cathode and zinc for the anode, with energy efficiency equal to or greater than lithium-ion batteries according to the company's description. It described local procurement of key raw materials such as halogen and zinc as a supply-chain advantage for Japan.

Those are company claims, so they need to be read as claims. The commercial test will come later: mass production, container-level density, cycle life, cost, serviceability, and performance under real data-center and grid conditions.

Still, the technology choice is notable. AI data centers concentrate large electrical loads. Operators want storage that can help with backup, peak management, grid flexibility, and possibly renewable integration. A safer aqueous chemistry is attractive if it can meet the density, economics, and reliability bar.

The Sakai plan puts storage beside AI hardware

Data Center Dynamics reported that SoftBank's telco arm will build large-scale battery cells and storage systems for the Japanese market, with production at the Sakai City factory that will also house a large AI data center and AI hardware plant. The Data Center Dynamics article on SoftBank's storage business said production is expected to start in 2028, with mass production targeted for 2029.

The Next Web described the project as a large-scale battery manufacturing effort at the former Sharp plant, with target output around one gigawatt-hour a year and partners Cosmos Lab and DeltaX. The TNW report on SoftBank's Sakai battery plant said the plant would supply storage for SoftBank's AI data centers and for grid, industrial, and residential customers beyond that.

That customer mix matters. If the plant only served SoftBank's own data centers, it would be a vertical-integration story. By pointing to grid, industrial, residential, and possible overseas markets, SoftBank is tying AI infrastructure to a broader storage market.

AI power demand is already testing grid planning

SoftBank's announcement lands while AI data-center demand is already forcing grid planners to update assumptions. Pagalishor's recent coverage of NERC's data-center grid alert focused on reliability risks created by large, fast-changing data-center loads. Another Pagalishor analysis of data centre renewable energy rules examined how policy makers are trying to tie new computing demand to clean power and grid flexibility.

The storage question sits between those two problems. Batteries cannot create generation by themselves. They can shift energy across hours, reduce peak stress, provide backup, smooth renewable output, and support local resilience if they are sized and operated correctly.

That makes the SoftBank battery business relevant beyond Japan. The company is effectively saying that AI buildouts need dedicated storage capacity and that storage may become part of the infrastructure stack, much like servers, networking, cooling, and land.

One gigawatt-hour is large but not a magic answer

A gigawatt-hour-scale target sounds big because it is big. It does not solve the AI power problem by itself.

One gigawatt-hour describes energy storage capacity, not continuous generation. A battery system can discharge stored energy for a period of time, but it still needs charging from the grid, solar, wind, gas, nuclear, or another source. For data centers that run around the clock, storage helps with resilience and flexibility; it does not replace firm power planning.

This is where AI infrastructure language can get sloppy. Batteries are sometimes discussed as if they directly power data centers in isolation. In practice, the value depends on the operating model. Are the batteries used for backup? Peak shaving? Grid services? Renewable matching? Local microgrid support? Industrial customer products? The answer changes the economics.

SoftBank's official release points to stable power supply and efficient energy use, not a promise that batteries alone will carry AI computing. That is the right framing.

Supply-chain control is part of the battery argument

SoftBank's battery plan also has a supply-chain angle. Lithium-ion battery supply chains are heavily shaped by Chinese refining, cell manufacturing, critical minerals, and global competition for battery materials. Zinc-halogen batteries may not remove every dependency, but SoftBank says local procurement of materials such as halogen and zinc can strengthen Japan's supply chain.

That is important for AI infrastructure because power equipment, transformers, batteries, chips, and cooling hardware are all becoming bottleneck categories. A company that owns more of its energy-storage supply may have more flexibility when data-center deployment schedules tighten.

Tom's Hardware reported that SoftBank is working with Cosmos Lab and DeltaX on water-based storage technologies and that the Sakai site is being shaped into a vertically integrated complex involving an AI data center, solar panel production, AI hardware, and batteries. The Tom's Hardware report on SoftBank's AI data-center batteries also noted the company's target of more than 100 billion yen in annual battery revenue by 2030.

That is a manufacturing bet as much as an energy bet.

Grid customers may care as much as data centers

SoftBank said its battery business will contribute to stable electricity supply and efficient energy utilization in Japan. That expands the audience beyond data-center operators.

Japan has a constrained power system, earthquake and disaster-resilience concerns, industrial demand, and pressure to integrate more renewable energy. Battery energy storage can help with local flexibility if it is deployed where grid operators, utilities, and large customers actually need it. The challenge is matching projects to grid conditions, not just announcing capacity.

The Register reported that SoftBank initially plans to deploy battery systems at its own large-scale AI server farms, with wider availability later. The Register article on SoftBank's battery plan said the company wants gigawatt-hour-per-year manufacturing capacity and expects the operation to generate more than $600 million a year by 2030.

That creates a two-stage test. First, can SoftBank make the batteries at scale? Second, can those batteries win customers outside SoftBank's own AI buildout?

The timeline gives planners a 2028 checkpoint

The useful date in the announcement is 2028. SoftBank plans to begin manufacturing battery cells and energy storage systems in the fiscal year ending March 2028 and aims for mass production around fiscal 2028. That gives grid planners, AI infrastructure teams, and industrial buyers a concrete checkpoint.

Between now and then, several things need to happen. SoftBank and its partners need to validate mass-production methods. The Sakai site needs to support the physical factory plan. Battery performance claims need to hold under commercial conditions. Customers need financing, contracts, warranties, safety approvals, and integration partners.

The gap between announcement and production matters because AI data-center demand is moving quickly. Utilities and regulators are already dealing with interconnection requests, load forecasts, and power-supply commitments. A 2028 storage factory may help with future flexibility, but it does not remove near-term grid pressure.

That is why the announcement should be read as a medium-term infrastructure signal, not a near-term fix.

Battery storage fits the AI power stack only with controls

Storage can become another layer in the AI power stack, but it needs operational discipline. Batteries have to be dispatched according to grid conditions, customer needs, warranty limits, safety rules, and market prices. A badly operated storage system can sit idle when needed or degrade faster than expected.

For an AI data center, that means the storage system should be planned with the facility's load profile, cooling demand, backup requirements, grid tariff, renewable contracts, and outage tolerance. It also needs cybersecurity controls because energy systems connected to data-center operations become part of the critical operating environment.

Pagalishor's earlier piece on data center battery storage and the AI power test made the same point from the grid side: batteries are useful, but they are a bridge and flexibility tool, not a replacement for careful generation and transmission planning.

SoftBank's plan strengthens that argument. Storage is moving closer to the center of AI infrastructure planning because the power problem is becoming central.

The business model depends on buyers beyond SoftBank

The storage plan becomes more credible if it can serve customers outside SoftBank's own AI buildout. Internal demand can help a factory reach early scale, but a durable battery business needs repeat buyers, service contracts, installation partners, and grid-facing use cases that work after the first data-center projects are supplied.

That is why the revenue target matters. A 100 billion yen annual battery business by fiscal 2030 would require more than a captive supply arrangement. It implies sales into industrial, grid, and possibly residential storage markets, where customers will compare safety, cost, density, delivery time, warranty terms, and integration support against established lithium-ion systems and newer long-duration storage options.

The Sakai cluster gives SoftBank a stronger starting point than a standalone factory because it can test storage against real AI infrastructure demand. But the wider market will judge the product differently. Utilities and industrial buyers will care about bankability, maintenance, safety certification, degradation curves, and whether the systems can be financed without unusual risk premiums.

The policy question is who pays for flexibility

Storage becomes politically sensitive when the costs move from a private data-center plan into the wider power system. If batteries are built only for one company's own uptime, the economics are mostly a corporate infrastructure decision. If the same batteries also support grid services, reduce peak pressure, or defer network upgrades, regulators and utilities will want to know how the benefits and costs are allocated.

That question is already showing up in data-center power debates. Utilities are asking large-load customers to bring new generation, accept curtailment, pay deposits, or show flexibility before they connect. Battery systems can help answer some of those requests, but only if dispatch rules are clear and the grid operator can count on the capacity when stress arrives.

SoftBank's plan is therefore a useful signal for policy makers. AI infrastructure developers may be willing to invest directly in storage when power availability becomes a growth limit. The public interest test is whether those investments reduce pressure on other customers or simply reserve scarce grid capacity for the largest buyers.

That will shape how similar projects are judged outside Japan. A battery factory attached to an AI data-center cluster looks more defensible when it can show grid value, local jobs, safer chemistry, and a clear path for industrial customers beyond the anchor tenant.

The real test comes before mass production

The SoftBank battery business announcement is important because it makes a practical point: AI data centers are not only a chip and server story. They are a power system story, and storage is becoming part of the competitive plan.

The hardest questions now sit between the press release and the factory floor. Can the zinc-halogen cells meet commercial expectations? Can SoftBank produce storage at scale by 2028? Can the systems serve both AI data centers and grid customers? Can they compete on cost, safety, density, warranty terms, and installation speed?

Those answers will decide whether Sakai becomes a model for AI-era power infrastructure or a useful but limited corporate supply project. Either way, the direction is clear: the companies building AI capacity are no longer waiting for the grid to solve every problem for them.