SoftBank's battery factory is the AI capex story most analysts are missing
Plan (drafter input)
SoftBank is converting a shuttered Sharp LCD factory in Osaka into battery cell manufacturing for AI data centers — targeting 1 GWh/year, lithium-free and cobalt-free, 100 billion yen in annual revenue by FY2030. The post's angle: this is what AI capex looks like when it starts going vertical. The semiconductor buildout is well-covered. The power and storage layer is the part most analysts are still underweighting. SoftBank is not trying to compete with CATL. It is trying to own the energy infrastructure its own data centers need. That is a different bet. Button: whether the zinc-halogen chemistry from Cosmos Lab actually scales is the technical question nobody in the AI capex conversation is asking yet.
Specific company, specific technology, specific numbers, and a genuine analytical angle about the AI capex story that extends Mia's stated beat on AI capex and whether it becomes a productivity story. Evergreen enough to not need the news hook but grounded in a real event.
special_message: Generate exactly 5 items: 1 with content_format='video' and 4 with content_format='hero_text'.
Body
The semiconductor buildout gets all the coverage. Every analyst has a slide on H100 allocations and data center square footage. The power and storage layer — the part that actually determines whether any of this runs — is still being underwritten like a footnote.
SoftBank's mobile unit is converting a shuttered Sharp LCD factory in Sakai, Osaka into battery cell manufacturing. Target: roughly 1 GWh per year initially, scaling from there. The cells are lithium-free and cobalt-free, developed with South Korean partners Cosmos Lab and DeltaX. Cosmos Lab brings the zinc-halogen chemistry; DeltaX brings the battery design. The revenue target is more than 100 billion yen from the battery segment by fiscal 2030. The organizing logic is explicit: SoftBank needs to power its own AI data centers, and it has decided that owning the storage layer beats buying it from someone else. That is a vertical integration bet, not a battery market bet. It is not trying to out-CATL CATL. Stationary storage for AI infrastructure has different requirements than EV energy density — it wants cheap, durable, and scalable, and the cobalt-free, lithium-free approach changes the cost structure and removes some supply-chain exposure that anybody sourcing from the DRC has learned to think about carefully.
The number worth watching is whether zinc-halogen chemistry from Cosmos Lab actually scales to several gigawatt-hours. That is the technical question the AI capex conversation is not asking yet. If the chemistry holds at scale, the model is interesting. If it doesn't, SoftBank has a very large converted LCD factory.
Caption
SoftBank is building its own batteries for its own AI data centers. The power layer is the part the capex conversation keeps skipping. #softbank #aicapex #energystorage #macro
Pipeline
- Hero image done fal · fal-ai/flux-pro/v1.1-ultrananfkXVI9xBi_hero.png$0.06api 18.3sMay 11, 3:07 PM
Chat References
No bot turns have referenced this post yet.