SoftBank is spinning up a new company called Roze AI that wants to use autonomous robots to build data centers, and it’s already lining up a $100 billion IPO before most people have even heard the name. According to TechCrunch AI, the Japanese conglomerate is targeting a public listing as early as the second half of 2026, with details first surfacing in the Financial Times and Wall Street Journal.
This is a bold move even by SoftBank standards. The pitch: as the AI infrastructure buildout accelerates across the U.S., the bottleneck isn’t just chips or power, it’s the actual physical construction of server farms. Roze AI plans to attack that bottleneck with robots.
What Roze AI Actually Is
TechCrunch AI reports that Roze would focus on making U.S. data center construction more “efficient” by deploying autonomous robots on building sites. Think of it as automating the people who build the buildings that house the AI.
Key points from the reporting:
- Target valuation: $100 billion, per the FT
- IPO timeline: Some SoftBank executives want it done by H2 2026
- Core product: Autonomous robotics for data center construction
- Status: Still being assembled, with internal skepticism already brewing
The FT notes that some inside SoftBank are uneasy about both the valuation and the timeline. That’s worth flagging. A $100B IPO for a company that hasn’t shipped anything publicly would be one of the most aggressive pre-revenue listings in recent memory.
Why This Matters
The AI infrastructure race has a physical world problem. Hyperscalers are pouring hundreds of billions into compute, but data centers still get built mostly the way warehouses got built 20 years ago. Skilled labor shortages, permitting delays, and construction timelines are real constraints on how fast capacity can come online.
If Roze AI works, it compresses that timeline. If robots can pour concrete, run cabling, install racks, and assemble cooling systems faster than human crews, then the limiting factor for AI infrastructure shifts from labor back to chips and power. That’s a meaningful unlock for the entire industry.
This is also part of a broader pattern. TechCrunch AI points out that Jeff Bezos recently co-founded Project Prometheus, a startup buying industrial firms to modernize them with AI. Different angle, same bet: the next decade of AI value isn’t only in models, it’s in applying automation to the physical economy.
The Skeptic’s View
SoftBank’s track record is mixed. The same firm that backed ARM and Alibaba also poured hundreds of millions into Zume, the AI-driven pizza delivery startup that collapsed in 2023. Masayoshi Son swings hard, and he doesn’t always connect.
A few things that should give investors pause:
- No public product yet. Construction robotics is hard. Boston Dynamics has been at it for over a decade and is still not building data centers.
- $100B valuation pre-IPO. That’s roughly the market cap of a top 100 U.S. public company. For comparison, that’s bigger than FedEx or Lockheed Martin.
- Internal doubts. When the FT is already quoting skeptics inside the company before the announcement, it’s a signal.
What to Watch Next
A few things will tell you whether Roze AI is real or hype:
- First customer announcement. Who actually agrees to let these robots on a job site?
- Robotics partnerships or acquisitions. SoftBank doesn’t have a deep robotics stack today. Where does the tech come from?
- Pilot timeline. A 2026 IPO requires a working demo well before then. Watch for site visits and footage.
- SoftBank’s own commentary. TechCrunch AI says it reached out for comment. SoftBank’s official framing will set expectations.
For practitioners building anything in the AI infrastructure stack, Roze AI is a useful signal. The capital is now flowing not just into the digital layer of AI but into the physical layer that supports it. Even if Roze stumbles, the thesis behind it is going to attract more entrants.
Full details at the original TechCrunch AI report.