SoftBank Just Started Building Its Own Power Grid for AI

Your GitHub stars won't save you if you're behind on tech trends.

That's why over 200K engineers read The Code to spot what's coming next.

Join 200k+ engineers

Forget chips. Forget models. Forget the parameter count wars. The AI race just took a hard left turn into something nobody on Twitter is talking about loudly enough: the people building the future of artificial intelligence are now in the business of building power plants and batteries. SoftBank just lit the fuse, and the explosion is going to be felt across every hyperscaler, every grid operator, and every utility company on Earth. Buckle up. This one is wild.

Today's biggest tech story is not a new AI model. It is not a new chip. It is a Japanese telecom company announcing that it is going to start manufacturing gigawatt-scale batteries in a converted LCD factory because the global electrical grid cannot keep up with what AI is doing to it. SoftBank Corp, the mobile arm of Masayoshi Son's empire, just launched a battery business in Japan targeting one gigawatt-hour of annual production capacity to start, with plans to scale to multiple gigawatt-hours per year. Annual revenue target by 2030: roughly 637 million dollars from this single new business unit. And the kicker: the entire reason this exists is because AI data centers have become so hungry for power that one of the largest companies in Japan decided it was easier to just start making its own batteries than to wait for the grid to catch up.

The factory site is the old Sharp LCD plant in Sakai City, Osaka. SoftBank bought the rights to it last year for roughly 100 billion yen, originally to convert it into an AI data center for running OpenAI agents. Now the plan has expanded dramatically. The site will house two new business units: the AX Factory, which will handle AI data center operations and the manufacturing of AI infrastructure hardware, and the GX Factory, which will produce next-generation batteries, solar panels, and related products. Mass production of battery cells is scheduled for fiscal 2028, which begins in April 2028. The first wave of batteries will be used inside SoftBank's own data centers. The second wave will be sold to the Japanese power grid, industrial customers, and eventually residential consumers.

The Register's breakdown of SoftBank's broader AI infrastructure push: https://www.theregister.com/on-prem/2026/05/11/softbank-bets-on-battery-building-to-back-bit-barns/

Nikkei Asia's detailed reporting on the Sakai conversion: https://asia.nikkei.com/business/energy/japan-s-softbank-enters-battery-business-to-power-ai-data-centers

Here is where the story gets technically interesting. SoftBank is not making the lithium-ion batteries you have in your phone or your Tesla. It is making zinc-halogen batteries, in partnership with two South Korean firms: Cosmos Lab, which has developed a novel zinc-halogen cell chemistry that uses pure water as the electrolyte, and DeltaX, which builds the energy storage system integration. The result is a battery cell that SoftBank claims is the world's first to integrate two next-generation battery technologies in a single product, using a halogen-based cathode and a zinc anode.

The headline feature is the one that should make every data center operator on the planet pay attention. These batteries are non-flammable. Lithium-ion batteries, for all their density and cost advantages, have a well-documented tendency to catch fire in spectacular and expensive ways. That fire risk is the single biggest reason why hyperscale data center operators have been reluctant to install massive battery banks directly inside server halls. They have to either build separate fire-rated structures or accept the risk of a thermal runaway event taking out millions of dollars of compute in a few seconds. Zinc-halogen cells using water as an electrolyte effectively eliminate that risk. They also avoid the cobalt and rare-earth supply chains that have been under heavy Chinese export-control pressure for the past two years. That is a geopolitical hedge worth billions on its own.

The Cell to Pack technology DeltaX is contributing reportedly delivers world-class energy density, and SoftBank will pair the physical hardware with its own AI-based energy management system that forecasts power demand and routes electricity in real time. The whole stack is vertically integrated: SoftBank generates the AI demand, manufactures the batteries that smooth that demand, runs the AI-powered software that decides when to charge and discharge, and operates the data centers that consume the resulting electricity. That kind of vertical integration is what hyperscalers have been quietly building toward for the last three years, and SoftBank just leapfrogged most of them by going all the way to the chemistry layer.

The Next Web's deeper economic analysis of the Sakai plant: https://thenextweb.com/news/oftbank-sakai-battery-plant-ai-data-centres

Data Center Dynamics on the full timeline: https://www.datacenterdynamics.com/en/news/softbank-launches-japanese-battery-storage-business-production-plant-to-house-ai-data-center/

Zoom out. SoftBank is now majority shareholder of Arm, the chip designer behind nearly every smartphone processor on Earth, which recently launched its first Arm-branded data center processor specifically targeting AI workloads. SoftBank also owns Ampere Computing, which makes Arm-based server chips. It is the largest single investor in OpenAI, having committed roughly 22.5 billion dollars before the end of 2025. It just announced plans for a massive 10 gigawatt data center campus on Department of Energy land in Ohio. And now it is making the batteries that will power all of it.

Stack those pieces together and SoftBank has assembled, in slightly less than 24 months, one of the most vertically integrated AI infrastructure stacks on the planet. Chips through Arm and Ampere. Models through OpenAI. Data centers in Japan and the United States. And now the batteries to power those data centers, the energy management AI to optimize those batteries, and the solar panels to feed those batteries. Every single layer of the stack is owned, controlled, or heavily influenced by a single Tokyo-based holding company.

The strategic implication is enormous. For years the conventional wisdom has been that the AI race would be won by whoever had the best models. Then it shifted to whoever had the most GPUs. Then to whoever had the most data center capacity. Today's announcement makes it clear that the next phase is whoever has the most reliable electricity, and the people who win that phase are going to be the ones who own their power infrastructure end to end. SoftBank is now the most aggressive bet in the world on that thesis. Watch what they build next.

Telecoms.com on the broader 100 billion yen AI batteries push: https://www.telecoms.com/ai/softbank-launches-100-billion-ai-batteries-business

Yahoo Finance on what this means for AI's power crisis: https://finance.yahoo.com/sectors/technology/articles/softbank-taking-ais-growing-power-143843323.html

A few threads to watch. First, watch the other hyperscalers respond. Amazon Web Services, Microsoft, Google, and Meta are all dealing with the same power crunch, and they have all been hesitant to vertically integrate this far down the stack. SoftBank has just changed the calculus by proving that a major operator can make this leap in roughly two years from a standing start. Expect to see at least one major U.S. hyperscaler announce a similar push by the end of 2026.

Second, watch the chemistry. If zinc-halogen cells with water electrolytes actually scale to gigawatt-hour production without losing performance, it will be one of the most important industrial breakthroughs of the decade. The chemistry has been studied for years and has always been promising in the lab. SoftBank scaling it to commercial volumes would change battery economics globally, not just for data centers.

Third, watch Japan. The country has been quietly positioning itself as the home of the non-China AI supply chain, and SoftBank's vertical integration play is the most aggressive expression of that strategy yet. Combined with Japan's broader push into semiconductor manufacturing and its existing dominance in robotics, the next five years could see Japan reclaim a level of technological relevance it has not had since the 1980s.

We will keep tracking this story and bring you the next chapter as it lands. The grid wars are just getting started.