Summary: The Financial Times broke the story on April 29, 2026: SoftBank is putting together a standalone company named Roze that would use autonomous robots to speed up US data center construction — and IPO preparations are already underway. The asset consolidation is taking shape too: energy, land, and infrastructure holdings SoftBank already controls could be folded in, and ABB Robotics — the unit SoftBank agreed to purchase in October 2025 — sits on the bundling list.
The Financial Times broke the story on April 29, 2026: SoftBank is putting together a standalone company named Roze that would use autonomous robots to speed up US data center construction — and IPO preparations are already underway. Some executives are aiming for a listing as soon as the second half of 2026, with a valuation target in the range of $100 billion. CNBC and the Wall Street Journal picked the story up afterward; SoftBank offered no comment.
The story turns on interlocking gears: AI cannot run without data centers, data centers cannot rise without construction crews, and skilled building labor is the input the American market has the hardest time supplying. Masayoshi Son’s fix is to close that circle — have robots erect the buildings AI lives in, then take the building business itself to the public markets. The wager is on robotics as the enabling layer for the whole AI buildout, not on any one robot product.
The Business Roze Is Meant to Run
By the FT and CNBC accounts, Roze would be an AI-plus-robotics firm whose main trade is putting up data centers, setting autonomous robots on US server-farm sites to raise construction efficiency. The asset consolidation is taking shape too: energy, land, and infrastructure holdings SoftBank already controls could be folded in, and ABB Robotics — the unit SoftBank agreed to purchase in October 2025 — sits on the bundling list.
How the $100 Billion Figure Holds Up
There is arithmetic behind the $100 billion tag: if Roze combines ABB Robotics with SoftBank’s energy and land assets, one listed company would hold the AI infrastructure supply chain end to end — the robots that raise the facilities, the power that runs them, the ground they occupy. The FT and CNBC each trace the plan to the same driving hand: Son himself.
2026’s busy AI listing calendar strengthens the nerve: with Cerebras back in the filing queue and a string of infrastructure names behind it, investor appetite for stocks that profit from AI demand is running high.
Inside Skepticism, With Zume as the Cautionary Tale
The believers are not universal. The FT states outright that SoftBank’s own ranks harbor doubts about both the valuation and the listing schedule, with Middle East conflict uncertainty named as another variable. TechCrunch supplies an uncomfortable flashback: SoftBank once committed hundreds of millions of dollars to Zume, the robot pizza venture that ended in bankruptcy in 2023. Putting robots to work in the physical world is ground SoftBank has already stumbled on once.
SoftBank stock shed 0.9% the day the reports appeared, though it remains up more than 18% since the start of the year.
Roze’s Slot in Son’s Wider Holdings
Set against the rest of Son’s portfolio, Roze reads less like a separate wager than the building crew for his AI empire. SoftBank’s commitments to OpenAI now exceed $30 billion; Stargate, run jointly with OpenAI and Oracle, carries a headline figure of $500 billion; and the group is erecting large data centers of its own, Ohio among them. The Vision Fund recorded a $2.4 billion gain in the December quarter, with OpenAI supplying most of the lift — the strategy is already showing up in the books. Every branch of the group wants more buildings, finished sooner, so “who does the building” stops being a purchasing question and becomes group-level strategy. Son’s answer: keep a builder in-house.
What Infrastructure Watchers Should Take Away
Three threads deserve tracking. First, if construction robots genuinely reach job sites, one of the most painful US data center bottlenecks — the shortage of skilled trades — gets partial relief, the opening gigawatt-scale timelines need. Second, Roze’s method of bundling robots with energy and land for a listing could become a template rival conglomerates copy; this current has company — Project Prometheus, co-founded by Jeff Bezos, likewise plans to acquire industrial companies and rework them with AI, as TechCrunch reported in March. The wealthy are converging on one conclusion: AI applied to heavy industry is the next margin pool. Third, watch the execution risk: the valuation is an ambition, the schedule is a wish, even insiders are still arguing, and the company has yet to make its official debut — reading Roze as a stress test of Son’s storytelling is safer than reading it as an accomplished fact.