Masayoshi Son walked into a meeting last week and told his executives he wants $100 billion for a company that doesn't exist yet.

The company has a name. It's called Roze. It has a plan to build data centers using autonomous robots. It has assets — three of them, bought separately and pasted together: ABB's robotics division, picked up for $5.4 billion. Ampere Computing, picked up for $6.5 billion. DigitalBridge, picked up for $3 billion. What it doesn't have is a product, a customer, or a single completed data center.

KPMG is already handling the IPO paperwork. Son wants Roze public by the end of this year. Some of his own executives think both the timeline and the valuation are nonsense, and they're telling reporters so on background.

This is the same Masayoshi Son who put roughly $30 billion into OpenAI, which then committed to spending hundreds of billions on the Stargate data center buildout — a $500 billion project he announced last year alongside Sam Altman and Larry Ellison in the Oval Office. SoftBank's own market cap sits at $197 billion. The stock is up 196% over the last twelve months. Son is sitting on enormous paper gains and an enormous cash hole. Roze is how he plans to plug the hole.

The pitch sounds clean. Bundle SoftBank's robotics bets, an AI chip company, and data center real estate into one entity, slap an "AI infrastructure" label on it, and let public markets pay for the next round. ABB Robotics is real. Ampere Computing makes real ARM server chips. DigitalBridge owns real data center capacity. The integration thesis — robots that build data centers for the AI those data centers will run — is technologically coherent. None of it has actually been built yet.

The Financial Times broke the story with a quote from a SoftBank insider calling the valuation "ambitious." That's the polite word. The honest version is that Son needs liquidity to fund his OpenAI promises, and the public market is the bank he's planning to rob.

Compare this to Arm, the last big SoftBank IPO. Arm had thirty years of history, a real product, billions in licensing revenue, and engineers in Cambridge who actually shipped silicon. Even with all of that, Arm's IPO disappointed Son's targets. Roze has none of those things. It has three logos on a slide deck and a story.

My Opinion

I'll be blunt. This is a financial maneuver pretending to be a tech company.

Roze is what happens when a holding company runs out of patience with private valuations. Son spent $14.9 billion buying robotics, chips, and real estate over the last eighteen months. He needs to mark those assets up to keep the SoftBank flywheel spinning, because he's contractually on the hook for tens of billions more to OpenAI. Bundling the assets, calling them "AI robotics infrastructure," and floating them at $100 billion does that mark-up in one move. The actual operating business? We'll figure that out after the IPO clears.

What bugs me is that this works. Stargate is a real project with real political backing. AI data center demand is genuinely bottlenecked by construction speed and labor cost. ABB makes the best industrial robots on the planet. The story is plausible enough that retail investors will buy it, institutional desks will model it, and Son will get his liquidity. By 2028, when nobody can find an actual Roze data center built by actual Roze robots, the stock will have already done its job for SoftBank's balance sheet.

The next twelve months will tell us whether Wall Street has any memory left after the WeWork fiasco. My bet: it doesn't, and the IPO prices above $80 billion.

Author: Yahor Kamarou (Mark) / www.humai.blog / 02 May 2026