SoftBank Spins Out Roze — a Robotics Company That Builds Data Centres, Eyes $100B IPO
SoftBank is spinning out a new company called Roze that combines robotics and data centre construction, with plans to list it on a US exchange at a target valuation of $100 billion. The Financial Times broke the story Thursday, citing people familiar with the matter. Reuters and TechCrunch corroborated independently.
The structure is novel. Roze’s premise is that the physical bottleneck in AI infrastructure — land, power, permitting, workforce, and construction speed — is as solvable with automation as the software stack. The company would deploy robots and automated systems to build data centres faster and cheaper than conventional construction, then operate them.
The Infrastructure Gap Roze Is Betting On
Demand for AI compute has outrun the pace at which data centres can be built. Sightline’s infrastructure analysis from earlier this month found that only 40% of announced AI data centre capacity will actually come online, with most projects stalling on power grid interconnection, permitting timelines, and skilled labour constraints. Roze’s pitch is a direct response to that finding.
Robotics-assisted construction is not new, but applying it specifically to hyperscale data centre buildout at this scale is. SoftBank’s existing portfolio includes Boston Dynamics, Arm, and a deep network of semiconductor and infrastructure investments — all of which are relevant supply-chain inputs to what Roze would need to operate.
The $100B Number in Context
A $100 billion target valuation would make Roze comparable to CoreWeave at its recent $8.5B investment-grade debt round, but positioned differently: CoreWeave is a GPU cloud operator. Roze would be the entity that builds the physical facilities those operators rent. The analogy is closer to a vertically integrated construction and operations company than a cloud provider.
At that valuation, the IPO would require investors to believe Roze can capture a meaningful share of a global AI data centre construction market that Omdia and others project will require trillions in spending through the decade. The robotics premium depends on whether automated construction actually runs faster and cheaper at scale — a question that remains undemonstrated at hyperscale.
SoftBank’s Bet on Physical AI
Masayoshi Son has spent the last eighteen months publicly arguing that artificial general intelligence is imminent and that the physical layer — robots, energy, and infrastructure — will be the next constraint after models. Roze is the clearest structural expression of that thesis. If AI capex continues at $130B+ per quarter, somebody has to build the boxes. SoftBank is betting it should be them, and betting early enough to go public on the expectation of demand rather than the proof of it.
The timeline for a formal IPO filing has not been announced. Bloomberg reported Thursday that shipments are planned to begin before year-end, with the company targeting 100,000 units in production by end-2027. SoftBank’s $100B figure, if achieved, would represent the largest infrastructure-adjacent IPO in the AI wave to date.