SoftBank Group is building a new standalone company called Roze, designed to deploy autonomous robots for data center construction, with plans to take the venture public on a U.S. exchange at a target valuation of $100 billion. The Financial Times first reported the plans in late April 2026, and both Reuters and Bloomberg subsequently confirmed key details through their own sourcing. SoftBank declined to comment on the reports.
If the IPO moves forward at that valuation, Roze would enter rare company. For comparison, Arm Holdings listed at roughly $54 billion in September 2023, and Meta Platforms debuted at about $104 billion in 2012. Only a handful of technology IPOs have ever crossed the $100 billion threshold.
Why data center construction is the target
The logic behind Roze starts with a straightforward problem: the world cannot build data centers fast enough to keep up with AI demand. Microsoft committed $80 billion to AI-ready data centers for its fiscal year 2025 alone. Meta earmarked $60 billion to $65 billion. Amazon pledged more than $100 billion. Yet construction timelines for large-scale facilities still stretch 18 to 36 months or longer, constrained by labor shortages, permitting delays, and fierce competition for electrical grid capacity.
SoftBank knows this bottleneck firsthand. The company is a founding partner in the Stargate joint venture alongside OpenAI and Oracle, a project announced in January 2025 with an initial commitment of $100 billion and ambitions to invest up to $500 billion in AI infrastructure across the United States. Building out that kind of capacity requires construction at a pace and scale that the industry has never attempted. Roze, as described in the reporting, would address that gap directly by fielding fleets of robots capable of handling repetitive, hazardous, or precision-intensive tasks on construction sites, from materials handling and structural assembly to inspection of cooling and electrical systems.
SoftBank’s robotics history
Roze would not be SoftBank’s first foray into robotics. The conglomerate acquired Boston Dynamics in 2017, gaining control of one of the world’s most recognized robotics brands, before selling an 80% stake to Hyundai Motor Group in 2021. SoftBank also developed Pepper, a humanoid robot designed for retail and hospitality settings, and has maintained investments in automation startups through its Vision Fund.
That track record is mixed. Boston Dynamics produced impressive demonstrations but struggled to find large commercial markets during SoftBank’s ownership. Pepper was discontinued for individual consumers. The Vision Fund’s broader portfolio included spectacular wins, most notably a $20 million early bet on Alibaba that returned tens of billions of dollars, alongside painful losses on companies like WeWork, whose valuation collapsed from $47 billion to under $10 billion before its eventual bankruptcy in 2023.
Masayoshi Son, SoftBank’s founder and CEO, has consistently signaled that he views artificial intelligence as the defining technology of the coming decades. In recent years, he has shifted the company’s center of gravity toward AI infrastructure, culminating in the Stargate partnership and the Arm Holdings IPO, which gave SoftBank a publicly traded semiconductor asset at the heart of the AI chip supply chain. Roze fits that trajectory: rather than simply funding AI software, SoftBank is moving to control the physical layer that makes AI possible.
What Roze would actually build
The specific technology behind Roze has not been publicly demonstrated. No prototypes, patent filings, or pilot project results have surfaced as of June 2026. That gap matters, because construction robotics is a field where the distance between concept and deployment is significant.
Several startups have already staked out portions of this market. Built Robotics, based in San Francisco, has developed autonomous heavy equipment for earthmoving and excavation. Dusty Robotics makes a robot that prints full-scale building layouts directly onto concrete floors, replacing manual measuring. Canvas, acquired by Dusty in 2023, automated drywall finishing. None of these companies, however, has attempted the kind of end-to-end construction automation that Roze’s description implies.
Data centers present both advantages and challenges for robotic construction. On the positive side, these facilities are highly standardized compared to residential or commercial buildings. Server halls, cooling systems, and power distribution units follow repeatable designs, which lends itself to automation. On the other hand, data center construction involves complex coordination across dozens of trades, strict safety and quality standards, and site-specific variables like soil conditions, seismic requirements, and utility connections that resist one-size-fits-all solutions.
The IPO timeline and its uncertainties
Bloomberg reported that the listing could slip from 2026 into 2027, and several factors could push it further. No S-1 registration statement or equivalent filing has appeared in SEC databases. The U.S. IPO market, while more active in 2026 than during the drought years of 2022 and 2023, remains sensitive to interest rate movements, geopolitical risk, and investor appetite for pre-revenue companies at high valuations.
SoftBank would also need to answer questions that the current reporting leaves open. How much revenue, if any, does Roze generate today? What is the leadership team? Will outside strategic investors, such as cloud providers or construction firms, participate before the IPO? And critically, will Roze consolidate any of SoftBank’s existing robotics or AI assets, or will it be built from scratch?
Institutional investors who lived through the SPAC boom and subsequent correction are likely to scrutinize a $100 billion pre-IPO valuation for a company with no public financials and no demonstrated product. SoftBank’s credibility on valuation has been tested before. The Vision Fund marked WeWork at $47 billion in early 2019; by the time WeWork finally went public via SPAC in 2021, its valuation had fallen to roughly $9 billion.
What Roze signals about the AI infrastructure race
Regardless of whether Roze reaches its IPO target on schedule, the venture reflects a shift in how the technology industry thinks about AI growth. For the past several years, the conversation has centered on chips, models, and software. Nvidia’s GPU dominance, OpenAI’s model releases, and the race to build larger training clusters have consumed most of the attention and capital.
But the physical constraints are catching up. Energy costs for data centers have surged. Utility companies in Virginia, Texas, and other data center hubs have pushed back on new connections, citing grid strain. Communities near proposed facilities have organized opposition over noise, water use, and environmental impact. Construction labor markets remain tight, with skilled electricians and mechanical contractors in especially short supply.
SoftBank’s bet with Roze is that robotics can break through at least some of these constraints, compressing build times and reducing dependence on scarce human labor. If the technology works at scale, it could give SoftBank and its partners a structural advantage in the race to deploy AI infrastructure. If it doesn’t, Roze risks becoming another ambitious SoftBank venture that promised transformation and delivered something considerably less.
For now, the most honest assessment is that Roze is a high-conviction bet from a company with a history of making them, backed by credible financial reporting but not yet by any public evidence of working technology. The next milestones to watch: a formal SoftBank announcement, any disclosure of Roze’s leadership or technical capabilities, and the eventual appearance of SEC filings that would make an IPO timeline concrete rather than speculative.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
