SoftBank forms robotics unit

Data centers are turning into the hard part of the AI boom. Chips matter, obviously, but the real slowdown now is often the building itself — land, power, permits, crews, and the sheer mess of getting giant sites finished on time. That is the gap SoftBank is trying to attack. The company is putting together a new business called Roze AI, aimed at using autonomous robots to make U.S. data-center construction faster, with an eye toward a U.S. IPO as soon as the second half of 2026. (techcrunch.com) ### What is Roze actually supposed to do? Roze is not just another software wrapper around AI infrastructure. The idea is a standalone robotics business focused on the physical work of building server farms — the repetitive, dangerous, and labor-hungry tasks that slow large construction projects (techcrunch.com) fits SoftBank’s broader push into AI chips, energy, robotics, and data centers as one connected stack. (techcrunch.com) ### Why target construction, not the servers? Because the bottleneck has moved. For a while, the story was mostly Nvidia GPUs and who could get them. But hyperscalers and AI labs are now running into a more old-school problem: you cannot deploy compute if the building is late. One recent FT repor(techcrunch.com)nal that construction itself is now strategic. (ft.com) ### Why does robotics help here? Construction is still weirdly manual for something this capital-intensive. Data centers are also more repeatable than many other building types — rows of racks, power systems, cooling gear, and standardized layouts. That makes them a better target for automation than, say, a custom office tower. The promise is not that robots replace eve(ft.com)obs so projects can move with fewer labor bottlenecks and less schedule slippage. That is the pitch behind Roze. (techcrunch.com) ### Why is the $100 billion number such a big deal? Because it tells you SoftBank is not treating this as a side project. Reports say Masayoshi Son wants Roze listed in the U.S. at a valuation around $100 billion, potentially as soon as late 2026. That is an enormous number for a company built ar(techcrunch.com)y, SoftBank wants investors to price robotic construction as core AI infrastructure, not as a niche industrial tool. (markets.ft.com) ### How does this connect to Stargate? Pretty directly. SoftBank is already tied to OpenAI and Oracle in the Stargate buildout, the huge U.S. AI-infrastructure push announced in January 2025. If SoftBank believes the next constraint is no longer financing or chips but the speed of deployment, then owning a robotics layer for construction makes strategic sen(markets.ft.com)ical rollout of the buildings those models need. (techcrunch.com) ### What is the catch? Robotics in construction is much harder than robotics in a factory. Factories are controlled. Construction sites are chaotic. Weather changes, layouts shift, and human crews improvise constantly. So the near-term win is likely narrow automation, not fully robotic building sites. That still could matt(techcrunch.com)l work will allow. (ig.ft.com) ### Bottom line SoftBank is making a very specific bet: the next valuable AI company may not train models or design chips, but remove the physical friction that keeps data centers from coming online. Roze is that bet in company form — and the IPO ambition shows SoftBank thinks the market may buy it. (techcrunch.com)o/))