$100 billion.
That is the valuation SoftBank is reportedly targeting for a new entity that does not yet exist—one that would build the very data centers needed to power the AI revolution Son believes is inevitable. The disclosed plans, first reported by TechCrunch, confirm that SoftBank is creating a standalone robotics company with a singular focus: constructing AI infrastructure at a scale that traditional construction firms cannot match.
The circular logic here is not lost on anyone who has watched Son recalibrate his investment thesis over the past three years. WeWork represented a bet on digital real estate. Arm Holdings was a bet on chip architecture. This new venture is a bet on the physical layer of AI—the concrete, steel, and wiring that no model can replace but every deployment requires. Son is not merely funding robotics companies anymore. He is funding the infrastructure that robots build, for the AI systems that justify building it.
The investor thesis beneath this move is straightforward: compute scarcity is the binding constraint on AI deployment, and the construction of data centers is the binding constraint on compute. Whoever controls that bottleneck controls the terms of the entire industry. If SoftBank can position its new entity as a preferred builder for hyperscalers and AI labs—guaranteeing delivery timelines that no contractor can match today—the $100B valuation becomes a multiple on a captive customer base, not a speculative fantasy.
The market context reinforces this urgency. NVIDIA's Blackwell chips require specialized cooling and power infrastructure that conventional data center designs cannot accommodate. The gap between chip availability and floor-ready capacity has stretched to eighteen months in some regions. A robotics firm purpose-built to close that gap addresses a real bottleneck, not an imagined one.
What remains uncertain is execution. SoftBank has a history of ambitious infrastructure announcements that outpace delivery timelines. The Vision Fund poured capital into companies that promised to reshape physical industries—Didi, WeWork, Oyo—and the outcomes were mixed at best. Building data centers at the scale Son envisions requires land, power procurement agreements, and construction expertise that no robotics startup has assembled from scratch.
But the difference this time may be the depth of vertical integration Son is proposing. The robotics division does not merely install pre-made servers; it fabricates the physical shell, installs the cooling systems, and potentially maintains the facility using autonomous systems. If that loop closes, the entity becomes simultaneously a customer for AI services and a provider of the infrastructure those services require. The $100B IPO target is not just a fundraising number. It is a statement that Son intends to own a node in the AI supply chain that cannot be easily replicated or disrupted.
The market will price this thesis against SoftBank's balance sheet and Son's track record. For now, the $100B number serves its purpose: it forces everyone to take the bet seriously, even those who have learned to discount SoftBank's more theatrical announcements. The physical AI infrastructure thesis is coherent. Whether SoftBank can execute it at scale is the only question that ultimately matters.