Robotic Lifestyle

ABB Group to sell ABB Robotics to SoftBank for $5.375B - The Robot Report
Humanoids·3 min read

SoftBank, NVIDIA stacks and the missing links: Where humanoid robots really stand in 2025

By Sophia Chen

SoftBank’s $5.375 billion bid for ABB Robotics and a flurry of platform launches from NVIDIA partners are colliding with hard engineering realities: power, sensing, and safe interaction. The next 12–18 months will test whether capital and edge‑AI stacks can push humanoids from lab demos into repeatable industrial deployments.

SoftBank’s $5.375 billion bid for ABB Robotics and a flurry of platform launches from NVIDIA partners are colliding with hard engineering realities: power, sensing, and safe interaction. The next 12–18 months will test whether capital and edge‑AI stacks can push humanoids from lab demos into repeatable industrial deployments.

Why this matters now: On Oct. 8, 2025, ABB Group agreed to divest its Robotics & Discrete Automation unit to SoftBank for $5.375 billion — a bet Masayoshi Son described as moving into “physical AI.” At the same time, vendors from Taiwan to Israel are shipping edge AI platforms and wireless power systems explicitly positioning themselves to service mobile manipulators and humanoid‑class machines. Those two forces — big capital plus specialized infrastructure — set up a rare inflection in humanoid development.

SoftBank’s ABB buy: scale, software, or a strategic mismatch?

SoftBank’s ABB buy: scale, software, or a strategic mismatch?

The numbers are blunt. ABB Robotics posted $2.3 billion in 2024 revenue and employs roughly 7,000 people; SoftBank’s offer values the division at $5.375 billion and aims to close in mid‑to‑late 2026 (subject to approvals). SoftBank brings deep pockets and a history of buying into robotics — from Aldebaran (Pepper) to Boston Dynamics — and Masayoshi Son framed the deal as a fusion of “artificial super intelligence and robotics.” (ABB sale announcement: https://www.therobotreport.com/abb-group-sells-abb-robotics-softbank-5-3b/).

Edge AI, sensor fusion and the software stack — small wins that add up

For humanoids, ABB contributes industrial‑grade actuation, the OmniCore controller family, and software tools proven across factories. That hardware maturity is critical: humanoids need high‑bandwidth motor control, deterministic safety layers, and field‑hardened perception pipelines to operate around people. The risk is organizational: ABB’s stack is optimized for repeatable, safety‑certified industrial arms and AMRs, not legged balance, whole‑body compliance, or dynamic bipedal locomotion — areas where TRLs remain in the mid‑range (TRL 4–6) for most humanoid firms.

Edge AI, sensor fusion and the software stack — small wins that add up

Power, uptime and the unsung infrastructure that enables humanoids

On the software side, companies such as YUAN are rolling out Jetson Orin–based platforms, Isaac ROS integration and NVBLOX mapping tools that reduce perception latency and enable multi‑sensor fusion on the edge. YUAN’s ARC AI Platform specifically claims ambitions for next‑gen humanoid intelligence by leveraging NVIDIA’s Isaac Sim and Jetson Orin family for real‑time decisioning (YUAN announcement: https://www.therobotreport.com/yuan-unveils-next-gen-ai-robotics-powered-by-nvidia-for-land-sea-air/).

Practical implications: reduced inference latency (tens of milliseconds) lets a humanoid close the loop for reflexive balance corrections and reactive grasping. But software alone won’t bridge the gap. Whole‑body controllers still struggle with robust recovery from unexpected contacts, and perception pipelines must handle cluttered, dynamic human environments with sub‑meter accuracy and strong failure‑mode detection. That’s a tough software engineering problem that requires large datasets, synthetic training in simulation (Isaac Sim), and rigorous field validation.

Who wins, who loses and the safety bar that can’t be skipped

Power, uptime and the unsung infrastructure that enables humanoids

One of the most overlooked bottlenecks for humanoid adoption is continuous power delivery. CaPow’s Genesis power‑in‑motion system won CE approval in early October and claims to eliminate downtime for AMRs, shrinking fleet size by up to 30% and enabling 100% uptime in some trials (CaPow CE announcement: https://www.therobotreport.com/capow-genesis-power-in-motion-tech-receives-ce-mark-european-market/). CaPow’s CEO Mor M. Peretz told reporters that their modular antennas install in 10–20 minutes and can support higher‑power mobile manipulators — a capability that could materially improve humanoid utilization.

Why that’s consequential: humanoids are power‑hungry because legged locomotion and dexterous manipulation both require high peak currents and sustained energy. Without a reliable charging and energy‑distribution strategy, deployments become logistics problems. Wireless floor infrastructure, combined with better battery chemistries and on‑board power management, could raise operating utilization from lab‑like bursts to continuous shifts — a necessary step to reach the ROI thresholds industrial buyers demand.

Sources