SoftBank is preparing a US IPO for a new venture called Roze — a robotics company whose stated job is to automate the construction of AI data centers — at a target valuation of roughly $100 billion. The Financial Times broke the story this week, citing executives close to Masayoshi Son. KPMG has been hired to prepare the financials. An analyst day is being planned for July at a Texas data center site to drum up demand. The IPO is targeting the second half of 2026.
Roze hasn't announced a product yet. It hasn't disclosed a revenue plan. It's a $100 billion company that does not yet exist as a company in the conventional sense. Even some inside SoftBank are reportedly skeptical of the valuation.
And yet the structure of the bet is so on-brand for Masa Son that you can almost hear the thesis being read aloud.
The Recursion
Here is what Roze is, in one sentence: a robotics company that uses AI to build the data centers that train the AI that runs the robotics.
That sentence is either a generational venture thesis or a perfect parody of one, and the answer probably depends on whether you're long or short SoftBank. Either way, it is the cleanest articulation of what's happening across the AI infrastructure economy in the spring of 2026: compute is the bottleneck, data centers are the product that solves the bottleneck, and the only way to build data centers fast enough is with the same technology you're trying to power.
KKR raised over $10 billion this week for a similar play — Helix Digital Infrastructure, dedicated to AI data centers and the power generation that feeds them. Bezos is funding Project Prometheus. Stargate, the joint OpenAI/SoftBank venture, has committed to spending in the hundreds of billions on infrastructure. Roze is not a wild idea. Roze is the Masa Son version of an idea that has now become consensus.
What's In The Box
The known assets bundled into Roze include ABB Robotics — the Swiss-Swedish industrial robotics giant that SoftBank agreed to acquire last year — and a portfolio of energy, land, and infrastructure holdings already in the SoftBank corporate parent. ABB is the operationally serious piece. It builds the robotic arms and assembly automation that run actual factories today. Pairing that hardware with AI software trained on construction data is, on paper, a real product wedge. The physical-AI layer Roze is betting on runs on purpose-built silicon, and the chip industry is consolidating to supply it — onsemi agreed to acquire Synaptics for $7 billion to build out edge-AI and sensing chips for exactly these robotics and IoT workloads. The private market has since put a price on that layer directly: Walden Robotics, a Toyota Research Institute and MIT spin-out, launched from stealth in July with a $300 million seed at a $1.1 billion valuation co-led by Toyota, on the strength of robots that have run eight-hour shifts at a Toyota plant in North America since February — industrial deployment first, valuation second, which is the inverse of Roze's sequence.
The pitch is that data center construction in the US is currently compute-constrained, labor-constrained, and time-constrained all at once. NVIDIA is trying to ship more H300s and Blackwells. Hyperscalers are trying to absorb them. Local governments are trying to permit power lines fast enough to feed the buildings. Every step of the chain is bottlenecked. Roze's pitch is that automating the build itself — site prep, structural assembly, equipment installation — collapses one of those constraints from years to months.
If that pitch lands, $100 billion is reasonable for a company with ABB's revenue base and a credible AI-construction story. If the pitch is just narrative scaffolding, $100 billion is a Vision Fund-style markup looking for an exit.
Why Now
The IPO window is the giveaway. SoftBank just committed an additional $30 billion to OpenAI at the start of the year, funded by a $40 billion bridge loan. Vision Fund 2 is winding down. The parent company needs liquid markups and exit ramps. A US IPO of Roze in the back half of 2026 — riding the wave of AI infrastructure euphoria — would generate the kind of paper gain that lets Masa keep playing the game he's been playing since the dot-com era.
This is the part of the cycle Masa is best at. Identify a thesis just as it's becoming consensus. Bundle pre-existing assets into a new corporate vehicle. Brand it aggressively. List it on US markets at a number that makes the original asset cost look quaint. Use the proceeds to fund the next bet.
The pattern worked spectacularly with Arm in 2023. It worked less spectacularly, in retrospect, with WeWork in 2019. Roze sits somewhere in between, with two crucial distinctions: ABB Robotics is a real industrial business, not a real estate sublease arbitrage; and AI data center demand right now is structurally under-supplied, not over-hyped.
The Tell
Watch the analyst day in July. SoftBank doesn't host analyst days at Texas data centers because Masa loves Texas barbecue. He hosts them when he needs institutional money — pension funds, sovereign wealth, Tier 1 mutual funds — to walk a site, see a robot move dirt, and price an IPO based on what they saw with their own eyes. The visual is the pitch. The valuation flows from the visual.
If Roze prices at $100 billion, the AI infrastructure top is further away than it looks. If it prices at $40 billion, the consensus has already started to discount the recursion. Either way, the August headline is going to tell us something about how much narrative pricing power AI still carries — and Masa is the perfect test pilot to find out.
Related on Uristocrat: OpenAI's $110B funding round, and Agility Robotics' $2.5 billion SPAC listing — the first pure-play humanoid robotics company to go public.
Comments