Artificial Intelligence
SoftBank Roze IPO: AI Data Center Robotics Bet
When it comes to investing early in tech trends, very few companies can be compared to SoftBank, the Japanese investment fund founded by Masayoshi Son. For example, Softbank has been an early investor in Alibaba (BABA ), Nvidia (NVDA ), Arm Holdings (ARM ), and OpenAI.
The company is now looking at locking in the valuation of another investment, Roze, an AI and robotic company automating the construction of AI data centers.
It should not be confused with RoseAI, an AI company for the finance sector, nor with RozeAI, a South Korean AI disaster solution company. The risk of confusion is especially high as RozeAI is planning a public IPO in May 2026 under the ticker ROZE.
Softbank’s Roze will be publicly listed sometime soon, ideally as early as 2026. This could represent one more major win for Softbank (SFTBY) and a reason for investors to pay attention again to the company’s stock.
Data Center Building Ramping Up
As AIs’ performances exploded in the past years, so has the activity in building “hyperscale” AI data centers able to deliver the computing power to train and then apply these AIs to commercial applications.
Just the construction of data centers, not their operations, was a massive market in 2025 of $261B, expected to grow further 12.7% CAGR to $662B.
Source: GrandView Research
AI workloads, which accounted for roughly 25% of demand in 2025, are projected to represent 50% to 70% of all data center computing by 2030. In total, $3.7T to $8T could be spent in AI data center infrastructures in the next five years.
However, many data center construction projects are being delayed, which could derail these projections. Some of the delays are linked to issues with energy supply, grid capacity, and, overall, the massive energy demand of these facilities.
Another issue is qualified manpower, including electricians, HVAC technicians, and plumbers. The problem compounds for data center-specific skills like superintendents, mechanical, electric, and plumbing coordinators; building information modeling (BIM) leaders; and commissioning managers.
Source: ThinkBRG
Over 60% of data center providers report challenges in finding qualified candidates for open roles.
”That is a problem when almost 3,000 data center projects are under construction or planned nationwide, including many large-scale buildouts, each of which typically employs between 1,500 and 3,000 workers during peak construction phases. Larger projects may employ as many as 4,000 workers, the equivalent size of a small town.”
This is why Softbank, at the forefront of this data center build-up with the $500B project Stargate, alongside OpenAI and Oracle (ORCL ), is looking to leverage robotics and AI to help build the AI data center capacity with the soon-to-be independently listed Roze company.
How Will Roze Be Built?
Creating A New Data Center Construction Method
Currently, Roze does not formally exist as a stand-alone company, nor does it have even a website of its own. However, it will be built out of the merger of several complementary companies and divisions fully or partially acquired by Softbank for several billion dollars:
- ABB’s robotics division was acquired for $5.4B.
- Ampere Computing was acquired for $6.5B.
- DigitalBridge data center assets, acquired for $3B.
The resulting Roze IPO is reportedly targeting a valuation of $100B, which would be a massive increase in value from the synergy of Softbank bringing together these separate capacities and leveraging its deep connection into the AI hyperscaler ecosystem, notably OpenAI.
It is not clear if other segments of Softbank will be integrated into Roze as well. A possible candidate is SB Energy, a SoftBank Group company focused on delivering energy supply systems to data centers.
Building Synergies
With the assets announced to be part of Roze, a model for the future new division of Softbank can be guessed at.
DigitalBridge brings its preexisting expertise as an asset manager dedicated to investing in digital infrastructure. The company has experience in the core business of Roze, data center, but also cell towers, fiber networks, and edge infrastructure.
We can expect DigitalBridge to bring the knowledge and human resources required for speeding up the designing, commissioning, and managing of AI data center construction projects.
“DigitalBridge is a leader in digital infrastructure, and this acquisition will strengthen the foundation for next-generation AI data centers, advance our vision to become a leading ASI platform provider, and help unlock breakthroughs that move humanity forward.”
Ampere Computing is a chip designer with around 1,000 semiconductor engineers who “design high-performance, energy-efficient processors specialized for next-generation cloud computing and AI workloads“. Together with Arm Holding, this segment will provide the computing hardware and chip expertise to speed AI deployment.
ABB robotics will be the part in charge of solving labor shortages, by replacing human labor or improving the productivity of workers when it comes to building, pulling cables, connecting pipes, installing server racks, etc.
What Will Make Roze Unique?
It is likely that it is ABB’s robotics division that will make Roze unique. ABB has very aggressively integrated AI into its industrial robots for many applications; for example, among the already deployed use cases that can be mentioned:
- Quality inspection: machine vision-enabled robotic cell to check spot and identify defects with 22-micrometer accuracy and at a speed twenty times faster than humans.
- Predictive maintenance: robots with AI-based assessment criteria learning delivers 99.9% diagnosis accuracy and zero production interruption.
- Autonomous drilling: robots with sensors and an AI system help Schindler with drilling and installation in the elevator shaft to improve precision and work safety.
- Item picking: robots can pick up unknown objects with the ability to execute 1400 picks per hour, powered by 3D vision and AI deep neural networks.
- Autonomous mobility: mobile robots with AI-enabled 3D vision navigation technology are able to work autonomously in a rapidly changing environment.
The fact that ABB has been focused for decades on heavy-duty industrial robots puts it in another category than most other “trendy” robotic companies, which are more focused on humanoid robots and other lighter models like robodogs.
Ultimately, the idea will be to build around the clock, 24/7, greatly speeding up the pace of construction.
This is part of a general trend of “Physical AI” becoming as much if not more important than pure software AIs (follow the link for a longer analysis on this topic).
Now, ABB Robotics will be benefiting from direct access to Softbank’s own AI systems, and also even more privileged access to the best AI models from Softbank’s closest partners, including OpenAI and Nvidia’s Omniverse-based RobotStudio HyperReality.
“Integrating NVIDIA Omniverse libraries into RobotStudio brings advanced simulation and accelerated computing to ABB Robotics’ unique virtual controller technology, accelerating how manufacturers of all sizes bring complex products to market.”
Deepu Talla – Vice president of robotics and edge AI at NVIDIA
The robot models likely to have the larger impacts are:
- IRB 8700, a 1,000 kg payload capacity robot, and the IRB 6700, for on-site automated welding, both robots are used for the heavier construction tasks.
- GoFa (CRB 15000)for automated screw-driving and installing sensitive electronic components, safe enough to be around human workers in the cramped aisles of a data center.
Source: ABB
How Will Roze Plans Work?
A likely element will be to make the data centers built by Roze as modular as possible and leverage off-site fabrication. ABB is already an expert in modular construction, where robots assemble timber or steel frames in micro-factories.
These pre-assembled modules will then be carried on the data center construction site, with potentially the construction of a segment taking as little as 12-24 hours, instead of days or even weeks.
Another element will be the large-scale use of digital twin technology. A copy of the data center in construction will exist digitally, and help the robots autonomously navigate the site to determine what goes where, and how.
It is also likely that later down the road, ABB robots will also be used for automated disassembly and recycling of server components, addressing the massive e-waste challenges of rapid AI scaling, with most AI data centers built today likely to be quite obsolete in a mere 5 years.
The Future of Roze Stock Post-IPO
For investors, the exposure to Roze can be achieved in two ways.
The first one is to own Softbank, and at the same time get exposure to the complex web of investments the company has built over the last few years, with many bold and aggressive bets on the rapid growth of the AI sector.
The second one is to wait for the IPO itself. Roze’s initial valuation is not yet official, but the rumoured $100B is considered by some as a little too high, considering that the automated robotic construction of data centers is far from an established practice yet.
On one hand, this is a risk as this means that there is little competition to compare it to, and the real market might be a lot smaller than expected, especially if the final performance in real construction sites is not as great as expected.
Potentially, more information from the IPO prospectus will clarify this topic a bit better. And of course, IPO valuation can indeed sometimes be overly optimistic, leading to a crash in the first weeks of trading.
On the other hand, being the very first company to automate, speed up, and overall improve the construction of data centers with robots is exactly the type of long-shot bet that has historically made SoftBank successful.
In addition, the same methods could be deployed for many other reindustrialization projects, from SMR power plants to semiconductor fabs, automotive plants, battery factories, or solar power plants.
It could be making Roze potentially the largest and highest-profile “pick and shovel” stock to invest in to get exposure to the AI buildup and the reindustrialization trend at large. And if the activity in the market over the past years has demonstrated anything, it is that the appetite for more major AI-centric stocks seems almost unlimited so far.
