The most honest sentence in tech right now comes from a TechCrunch headline: 'You need infrastructure to build AI, and robots to build infrastructure.' SoftBank is creating a company where robots construct the data centers that train AI that will eventually improve the robots. A $100B IPO target for a company that does not yet exist is either visionary or a Ponzi diagram. Possibly both.
The Ouroboros Economy and Agentic Capital
This is not isolated megalomania. A 2026 paper on arXiv, "Operating-Layer Controls for Onchain Language-Model Agents Under Real Capital" by Barton et al., examines autonomous language-model agents that translate user mandates into validated transactions under real financial stakes. The paper is essentially a safety manual for the thing SoftBank is building at civilizational scale. When agents control capital flows and also direct physical construction, the reliability question stops being academic. TurboFund's latest weekly signals note Sam Altman publicly calling for OS, UI, and internet protocols to be natively agent-compatible, which is the software layer that SoftBank's robot-built data centers would eventually run on. The infrastructure bet and the OS bet are convergent.
RAM, Robots, and the Physical Bottleneck
Meanwhile, Samsung is warning the RAM shortage gets worse through 2027, which means the fantasy of infinite scalable compute keeps hitting the very material reality of chips, fab capacity, and physical installation. Robots building data centers faster than humans could is not a gimmick. It is a direct response to a hardware supply chain that is genuinely constrained. The SoftBank bet reads less like hubris when you understand the shortage context. Whether the $100B valuation is justified is a separate question. Founders building in adjacent infrastructure should check TurboFund's list of 25 seed-stage AI investors as this sector heats up considerably.