Masayoshi Son just did something that makes his $40 billion OpenAI loan look like pocket change. SoftBank is spinning out a brand-new company called Roze — a U.S.-incorporated, U.S.-listed AI and robotics venture — and targeting a valuation of $100 billion in what would be one of the largest IPOs in history. Roze won’t write AI models. It won’t build chatbots. It will deploy autonomous robots to physically construct the data centers that every AI company on Earth desperately needs. And that distinction is the entire point.
The Man Who Lost $70 Billion on WeWork Now Wants $100 Billion for Robots
Let’s get the uncomfortable part out of the way. Masayoshi Son’s track record with mega-bets is, charitably, mixed. The Vision Fund’s spectacular implosions — WeWork, Katerra, Wirecard — cost SoftBank roughly $70 billion in write-downs. Son went quiet for two years. Then he came back swinging with a $40 billion loan to back OpenAI’s $500 billion Stargate initiative, and now he’s doubling down with Roze.
But here’s the thing: the market conditions that made WeWork a disaster are the exact conditions that make Roze potentially brilliant. WeWork needed infinite demand for office space that never materialized. Roze is betting on demand for AI data centers that is already so far ahead of supply that Big Tech is committing $700 billion this year alone trying to close the gap. Son isn’t betting on future demand this time. He’s betting on a bottleneck that already exists.
What Roze Actually Does — And Why It’s Not Just Another AI Company
According to reporting from CNBC, the Financial Times, and TechCrunch, Roze will be a U.S.-incorporated entity that combines three things SoftBank already owns or controls: ABB Robotics (one of the world’s leading industrial robotics firms, which SoftBank agreed to acquire last year), SoftBank’s existing energy and land assets earmarked for data center construction, and proprietary AI software to coordinate autonomous construction at scale.
The pitch is deceptively simple: instead of hiring tens of thousands of construction workers to build data centers — a process that currently takes 18 to 36 months per facility — Roze will deploy fleets of autonomous robots that can work 24/7, don’t need breaks, and can be reprogrammed for different site configurations. SoftBank claims this could cut construction timelines by 40 to 60 percent.
If that sounds like science fiction, consider this: ABB Robotics already has over 500,000 robots deployed globally in manufacturing, logistics, and assembly. The technology for robotic construction isn’t theoretical — it’s an integration challenge. And SoftBank is betting that bundling ABB’s hardware with AI coordination software is a $100 billion answer to the biggest infrastructure bottleneck in tech history.
Follow the Money: Why a U.S. Listing Changes Everything
The decision to incorporate and list Roze in the United States is the most strategically loaded part of this entire plan. SoftBank is a Japanese conglomerate. It could easily list Roze in Tokyo. Instead, Son is deliberately creating a U.S. entity that American institutional investors, pension funds, and sovereign wealth funds can buy without the friction of cross-border investment.
This is directly tied to the Stargate project. SoftBank committed $100 billion to Stargate — a joint venture with OpenAI, Oracle, and the Trump administration to build AI infrastructure across the United States. Roze is the construction arm of that vision. By listing it in the U.S., Son gets access to American capital markets to fund the very infrastructure the American government wants built. It’s a closed loop: U.S. policy demands AI data centers, SoftBank builds them with robots, U.S. investors fund the construction, and Son takes a cut of every facility.
The timing is not accidental. SoftBank’s stock rallied 7% on the Roze announcement alone. Son is riding the wave of AI infrastructure anxiety — the growing realization among investors that the $700 billion Big Tech is spending on AI this year has nowhere to go if the physical buildings can’t be built fast enough.
The $100 Billion Question: Is This Valuation Insane?
Several SoftBank executives are reportedly skeptical of the $100 billion target, according to the Financial Times. And they should be. Roze doesn’t have revenue yet. It doesn’t have a completed autonomous construction project to point to. It’s essentially a holding company for ABB Robotics plus some land and energy assets, wrapped in an AI narrative.
But valuation in 2026 isn’t about what you have — it’s about what you’re positioned to capture. Consider the math: if Big Tech spends $700 billion on AI infrastructure this year, and construction costs represent roughly 30 to 40 percent of total data center spend, that’s a $210 to $280 billion annual market for building these facilities. If Roze can capture even 5% of that market within three years, you’re looking at $10 to $14 billion in annual revenue from a market that’s growing 30%+ year over year.
At a 10x revenue multiple — generous but not unusual for a high-growth AI infrastructure play — that gets you to $100 billion. The math works if the robots work. That’s the bet.
Who Gets Hurt: The Second-Order Effects Nobody’s Discussing
If Roze succeeds, the ripple effects are enormous. Construction labor is the obvious casualty — data center construction employs hundreds of thousands of workers in the U.S. alone, and autonomous robotics could eliminate a significant portion of those jobs. But the deeper impact is on the existing data center construction firms — companies like Holder Construction, DPR Construction, and Hensel Phelps — that currently dominate the market. They’re suddenly competing with a $100 billion-funded robot army.
There’s also a geopolitical angle. If Roze can genuinely accelerate data center construction by 40 to 60 percent, it gives the United States a structural advantage in the AI infrastructure race against China. Beijing is building data centers as fast as it can, but it’s doing it with human labor. If American data centers start going up in 8 months instead of 24, the compute gap between the U.S. and China widens dramatically — and that’s exactly the outcome Washington wants.
The Verdict: Son’s Smartest Bet or His Most Dangerous One
Here’s where I land on this. The SoftBank Roze IPO is the single most important infrastructure bet in the AI era — more important than Stargate, more important than any hyperscaler’s capex commitment. Because all of that spending is meaningless if the buildings can’t be built fast enough. Son identified the chokepoint, bought the robotics company that could solve it, wrapped it in a U.S.-listed vehicle, and is now asking the market for $100 billion to execute.
The risk is real. ABB Robotics has never done autonomous construction at data center scale. The integration between robotics hardware and AI coordination software is unproven at this scope. And SoftBank’s internal skeptics are right that a $100 billion valuation for a pre-revenue entity is aggressive even by 2026 standards.
But the asymmetry is what matters. If Roze fails, SoftBank loses a chunk of capital — painful but survivable for a company with a $200 billion+ portfolio. If Roze succeeds, it becomes the default construction platform for every AI data center built in the Western world. And in a year where $700 billion is being spent on AI infrastructure with no end in sight, that’s a position worth betting on.
Masayoshi Son has been wrong before. But he’s never been wrong about a bottleneck this obvious.
