AI
SoftBank Roze AI IPO — Robots Will Build the Future of AI Infrastructu
SoftBank's Roze AI wants robots to build the data centers powering the AI boom. $100B valuation, 2026 IPO target, and a labor shortage that makes the problem real.
What Just Happened
The SoftBank Roze AI IPO is the most audacious bet in the physical AI race yet.
The AI infrastructure race has a problem nobody talks about enough. The United States is short roughly 439,000 construction workers. Satellite imagery suggests up to 40% of AI data center construction sites are already facing delays. Big Tech has committed over $650 billion in AI infrastructure spending in 2026 alone — and there aren't enough humans to build it fast enough.
Masayoshi Son thinks he has the answer. And he's betting $100 billion on it.
SoftBank is spinning out a new company called Roze — a U.S.-incorporated, U.S.-listed AI and robotics venture whose entire purpose is to automate the physical construction of data centers using autonomous robots. No more labor bottlenecks. No more construction delays. Just machines building the machines that run AI.
The Financial Times broke the story on April 29. The Wall Street Journal confirmed it. Reuters followed. Within 24 hours, every major outlet had the same details: Roze is real, it's targeting a $100 billion valuation, and SoftBank wants it public before the end of 2026. KPMG has already been hired to prepare the financials. An analyst day at a Texas data center is planned for July to drum up investor interest.
Son is calling this "physical AI" — the next frontier after software intelligence. The idea is simple and audacious: if AI is going to reshape the world, someone has to build the physical infrastructure that makes it possible. And if humans can't build it fast enough, robots will.
Whether the math works at $100 billion — for a company that hasn't shipped a single robot or completed a single data center — is the question the entire industry is now asking.
The answer depends on whether you believe Masayoshi Son has finally found the bet he was always building toward.
What Roze Actually Is
Roze isn't starting from nothing. That's the part of this story most coverage has missed.
SoftBank is bundling a set of existing assets into this new entity. ABB Robotics — one of the world's leading suppliers of industrial robotics and machine automation, which SoftBank agreed to acquire last year — is expected to be integrated into Roze, marrying world-class hardware with AI-driven software. Ampere Computing, DigitalBridge infrastructure assets, and SoftBank's existing energy and land holdings are also reportedly being consolidated under the Roze umbrella.
The core technology thesis is already in development. SoftBank has been experimenting with server rack designs that eliminate traditional cabling entirely, allowing autonomous robots to physically install and swap computing units by moving components into place — no human hands required. The goal is a fully robot-operated data center where construction, maintenance, and upgrades are handled end-to-end by machines. This is not a PowerPoint company. It's an assembly of real assets organized around a genuinely novel operational model. The question is whether that assembly is worth $100 billion before it has proven the model at scale.
The SoftBank Roze AI IPO and the Labor Wall Nobody Can Break
To understand why Roze exists, you need to understand the scale of the infrastructure problem Son is trying to solve.
The Stargate Project — the OpenAI-led consortium in which SoftBank is a key partner alongside Oracle and MGX — anticipates up to $500 billion in data center investment over four years. Microsoft, Google, Amazon, and Meta are collectively spending hundreds of billions more. The physical buildout required to support this level of AI infrastructure investment is staggering.
And the construction industry cannot keep up. The labor shortage is structural, not cyclical. Skilled electricians, data center technicians, and construction workers who can operate in high-density server environments are not produced fast enough by any training pipeline to meet current demand — let alone the demand of a $650 billion annual buildout.
Roze's pitch is that autonomous robots can break this bottleneck. If the technology works at scale, the total addressable market isn't just data center construction — it's every physical constraint on the global AI infrastructure buildout. That's the logic behind a $100 billion number for a company that doesn't yet have revenue.
The Valuation Problem
Even among SoftBank insiders, the $100 billion target has raised eyebrows. Multiple reports describe internal skepticism about both the valuation and the timeline. Geopolitical instability — specifically the ongoing situation in the Middle East — is cited as one factor that could delay the IPO or force a valuation reset.
The honest comparison is uncomfortable. Among companies that have actually completed public listings, only a handful in history have debuted at $100 billion — and all of them had massive existing revenue. Saudi Aramco. Alibaba. Facebook. Visa. Roze has none of that yet.
What Roze does have is Son's track record of using public markets strategically. His playbook with Arm Holdings — retaining roughly 90% of the company while listing it publicly to generate liquidity and market credibility — is the template here. The IPO isn't necessarily about raising cash. It's about establishing a public market valuation that SoftBank can use as collateral, as currency, and as a signal to the rest of the AI infrastructure market that Son controls a critical piece of the stack.
The Masayoshi Son Pattern
Son's career has a consistent shape. He makes bets that look insane at the time, gets most of them wrong, and occasionally gets one so spectacularly right that it defines a decade. His early investment in Alibaba — a $20 million bet that eventually returned over $70 billion — is the canonical example.
His Vision Fund era was messy. WeWork. Zume Pizza. A string of overpriced late-stage bets that burned billions when the easy money cycle ended. But his AI infrastructure thesis — heavy positions in Arm, OpenAI, and now Roze — reflects a different kind of conviction. He isn't betting on individual companies this time. He's betting on the physical layer of an inevitable buildout.
The difference matters. Software bets can go to zero. Infrastructure bets rarely do — because even when companies fail, the physical assets remain. Data centers, land, energy connections, and industrial robotics hardware don't disappear when a startup misses its revenue targets. They get acquired, repurposed, or sold.
What This Means for the AI Infrastructure Race
Roze signals something important about where the AI industry is heading. The software layer of AI — the models, the APIs, the applications — is increasingly competitive and increasingly commoditized. The physical layer — the power, the land, the compute, the buildings that house it all — is where durable advantage is being built right now.
Every major player in AI infrastructure understands this. Microsoft is spending $80 billion on data centers this year. Google is matching it. Amazon is building its own nuclear power deals to feed its compute appetite. Meta committed $115 billion in capex.
Son's insight with Roze is that none of these companies want to be in the construction business. They want infrastructure. If Roze can deliver it faster, cheaper, and more reliably than human-led construction, it becomes an essential vendor to every hyperscaler on earth simultaneously.
That's the $100 billion story. Not a robotics company. Not a data center company. A company that sits at the bottleneck of the entire AI economy and charges accordingly.
Whether the robots can actually do it — at the speed, reliability, and cost that makes the math work — is still unproven. But the problem Roze is trying to solve is unambiguously real. And in 2026, that might be enough to get it public.