SoftBank Targets $100 Billion IPO for Roze, the AI Robotics Company It Hasn't Built Yet

Masayoshi Son has made a career of placing enormous bets on imagined futures. SoftBank’s Vision Fund era was defined by pouring money into companies — WeWork, Uber, DoorDash — before their market dominance was established. Now Son is applying that same logic to an even more ambitious play: creating a company called Roze, targeting a $100 billion valuation and a U.S. IPO before the end of 2026, to build AI data centers using autonomous robots.

The catch: Roze doesn’t have a product yet. It may barely have a name.

What Roze Is Supposed to Be

As reported by the Financial Times and confirmed by multiple sources close to SoftBank, Roze would be a new spinout entity that bundles several of SoftBank’s existing AI infrastructure holdings into a single company with a unified strategic thesis: use robotics to make the construction and operation of AI data centers dramatically faster and cheaper.

The assets earmarked for inclusion paint a picture of how ambitious — and how expensive — the vision already is before a single line of product code has been written. The centerpiece is ABB Robotics, one of the world’s leading industrial automation and robotics suppliers, which SoftBank agreed to acquire for approximately $5.4 billion last year. Roze would also fold in Ampere Computing ($6.5 billion), a cloud-native Arm-based chip company, and DigitalBridge ($3 billion), a digital infrastructure investment platform. Additional energy assets, land holdings, and other infrastructure positions from SoftBank’s broader portfolio could also be included.

The strategic logic, at its core, is vertically integrated AI infrastructure: Ampere chips power the compute, DigitalBridge provides the infrastructure investment and operations layer, and ABB’s robots physically construct, wire, and maintain the facilities — ideally without the labor bottlenecks and cost overruns that have plagued conventional hyperscale data center build-outs.

The $100 Billion Number

The $100 billion valuation target is the figure that has drawn the most scrutiny — and the most internal SoftBank skepticism. According to sources cited by the Financial Times, some SoftBank executives privately view both the valuation and the H2 2026 IPO timeline as overly ambitious, especially given geopolitical uncertainties affecting capital markets and the current interest rate environment.

To put the number in context: Roze would be targeting a valuation roughly equivalent to Snowflake at its 2020 peak, or AMD at the height of its datacenter GPU run-up. The difference is that both of those companies had years of revenue, products, and customer relationships. Roze, at this point, is a strategic thesis and a collection of recently acquired assets being reorganized under a new name.

What gives the number even a veneer of plausibility is the macro backdrop. Big Tech companies have collectively committed over $700 billion in AI infrastructure capital expenditures for 2026, with hyperscalers competing furiously for data center capacity. If Roze can credibly position itself as a provider of robotic-enabled AI infrastructure construction — cutting build timelines from years to months — the addressable market is genuinely enormous.

The Stargate Connection

SoftBank’s involvement in Roze cannot be understood without its role in the Stargate project — the $500 billion joint initiative announced in early 2025 between SoftBank, OpenAI, Oracle, and several other technology partners to build AI data center capacity across the United States. Stargate established SoftBank as a central player in AI infrastructure at the national scale, with Son committing SoftBank to deploy $100 billion in the U.S. over four years.

Roze appears to be Son’s vehicle for making that commitment operational. Rather than simply writing checks to data center contractors, SoftBank would own the robotics-enabled construction platform itself — capturing margin across the entire stack from land acquisition to computing capacity. The commercial logic is that Stargate’s own build-out demand could become Roze’s anchor customer base, reducing the company’s go-to-market risk while creating a vertically integrated infrastructure play that external hyperscalers would also want to access.

What Doesn’t Exist Yet

Despite the ambition — and the nine-figure asset base already assembled — Roze has several critical gaps that make the H2 2026 IPO timeline look stretched.

No product has been announced. The company has not demonstrated a robotic data center construction system at any scale. ABB Robotics brings industrial automation expertise, but adapting factory-floor robotics to the highly variable, physically complex work of data center construction is a distinct engineering challenge that has not been solved publicly.

No revenue plan has been shared. Roze has not articulated a commercial model — whether it would function as a contractor building facilities for others, an owner-operator leasing compute capacity, or a platform licensing its robotic systems. Each of these is a fundamentally different business with different capital requirements, margin profiles, and competitive dynamics.

No IPO date has been set. The H2 2026 timeline is a target, not a filing. The gap between “targeting a $100 billion IPO by the end of the year” and completing a successful public offering is considerable, particularly for a pre-revenue company in a rising-rate environment where growth-at-any-cost valuations have come under sustained pressure.

Why Son Might Pull It Off Anyway

Dismissing Roze as a pre-revenue fantasy would be missing the pattern that defines SoftBank’s approach to large-scale technology bets. Son has consistently been willing to commit capital before a market’s structure has crystallized — and has occasionally been right in ways that transformed industries.

The AI infrastructure buildout is real, capital-intensive, and urgently needed. If Roze can demonstrate even a partial proof of concept — one robotic data center facility, one committed hyperscaler contract — before filing, the narrative becomes substantially more defensible to public market investors who have spent the last two years watching AI companies trade at multiples that would have seemed absurd in 2022.

Son’s track record on the “build the platform, wait for the market” approach is mixed. For every ARM Holdings — held patiently and floated at a $60+ billion valuation — there is a WeWork. Roze’s outcome will likely depend on whether the robotics-to-data-center thesis can be demonstrated at scale before the IPO window opens or closes.

What to Watch

The next concrete milestones to watch: a formal announcement of Roze as a legal entity, any disclosed customer relationships or LOIs with hyperscalers, and early signals from Goldman Sachs, Morgan Stanley, or other potential underwriters about whether they are engaged on an offering. If any of those materialize before June, the H2 2026 timeline becomes credible. If not, the IPO is more likely a 2027 story — or a vision that gets folded back into SoftBank proper.