SoftBank can document the data-center buildout. The robot part is still mostly a pitch.

SoftBank is trying to sell investors a future where robots help build the physical backbone of the AI boom, even though the public proof for that robotic future is still hard to see.

According to Reuters, citing the Financial Times, SoftBank plans to create and list a U.S. company called Roze that would be involved in building data centers, with a target valuation of $100 billion and a possible listing as early as this year. That is a striking pitch because data centers are now some of the most expensive and urgent construction projects in tech. But the nearest public documentation for Roze right now is that Reuters report, and SoftBank declined to comment.

What SoftBank can already point to is more concrete. In an October 2025 press release, SoftBank said it had agreed to buy ABB's robotics business, a major industrial robot maker, for $5.375 billion, with the deal expected to close in mid-to-late 2026. In a separate September 2025 announcement, SoftBank said its AI strategy now spans AI chips, AI robots, AI data centers, and energy, and that it had already broken ground on an advanced data center design in Lordstown, Ohio.

That matters because the story here is not that SoftBank woke up one morning with a new robot company name. The story is that Masayoshi Son has been assembling the parts for a giant AI infrastructure machine: chip ambitions, power plans, data-center construction, and now a robotics pipeline. Reuters reported in November that SoftBank planned to invest up to $3 billion to overhaul a former electric-vehicle factory in Lordstown so it could produce modular data-center equipment for OpenAI sites. Modular data centers are prefabricated units built in pieces and installed faster than a fully custom site.

SoftBank's broader Stargate push is also real. In its September announcement, the company said Stargate had reached nearly seven gigawatts of planned capacity and more than $400 billion in planned investment over three years. Those are SoftBank numbers, not independent construction audits. They still describe a company spending heavily on the physical side of AI.

The leap comes when that very real buildout gets translated into a near-term story about autonomous robots building data centers. Reuters' report says Roze will be involved in building data centers. It does not, at least in the public version available now, lay out what robots Roze has, what tasks they would perform, whether those robots already exist in the field, or how much of the work would actually be automated.

That gap matters more than the company name. Plenty of AI infrastructure work is repetitive, expensive, and painful enough that automation sounds plausible on first hearing. Data centers need steel, cooling systems, electrical wiring, prefabricated modules, logistics, and a lot of labor coordinated on brutal timelines. A robot that could take over even part of that process would matter. But there is a long way between "this is a good use case for robotics" and "here is a documented robotics business that can justify a $100 billion public valuation."

SoftBank does have one obvious answer to that skepticism: ABB. If the acquisition closes on schedule, SoftBank will control one of the world's best-known industrial robotics businesses. That gives Roze, if it materializes, a much firmer industrial base than a slide-deck startup would have. The problem is timing. SoftBank said the ABB deal is expected to close in mid-to-late 2026. Reuters reported Roze could be listed as early as this year. The IPO dream is arriving before the full robotics integration.

There is also a simpler possibility. Roze may turn out to be less a robotics breakthrough than a financial wrapper around assets SoftBank already has or plans to have: Lordstown manufacturing, Stargate data-center demand, ABB robots, and Masayoshi Son's habit of bundling adjacent bets into one grand narrative. That would not make the company fake. It would make the robotics label more aspirational than operational, at least for now.

Even the Ohio factory story points in that direction. Reuters reported in August 2025 that Foxconn and SoftBank planned to make data-center equipment in Ohio for Stargate, and that Foxconn was selling the Lordstown site and machinery for $375 million. That looks like real industrial preparation. It does not, by itself, prove robot-built data centers. It proves SoftBank is serious about manufacturing the hardware around them.

For founders and operators watching the AI buildout, that distinction is the whole story. SoftBank's infrastructure ambition is easy to document. SoftBank's robot-built data-center future is still mostly a promise attached to that ambition. Until Roze exists in public with products, customers, or machines doing visible work, the safer reading is that SoftBank is trying to IPO an automation narrative before it can fully show the automation. That's not nothing. But it is not the same as proof.