When SoftBank agreed to buy ABB Robotics for $5.375 billion, it signaled a new phase: industrial-robot expertise migrating into a physical-AI playbook that includes humanoids. The deal exposes an uncomfortable truth — hardware, control and industrial scale are proven; continuous power, safe autonomy, and regulatory readiness are not.

Why this matters now: SoftBank’s purchase (announced October 2025) of a unit that generated $2.3 billion in 2024 revenue and employs roughly 7,000 people shifts strategy from lab demos to commercial rollouts. Masayoshi Son framed the acquisition as an advance into “physical AI,” but moving from industrial arms to human-form robots requires different engineering, safety regimes and supply chains. (ABB expects the transaction to close mid-to-late 2026: https://www.therobotreport.com/abb-group-sells-abb-robotics-softbank-5-3b/.)

Power and uptime: the silent bottleneck

The stakes are concrete. Investors and integrators want humanoids that can operate 8–24 hours in real facilities, plug into existing workflows, and pass regulatory and safety audits. Today’s bottlenecks are power (runtime and charging logistics), robust perception and whole-body control at scale, and a tested path from prototype TRL levels to 7–9 for production use. Companies from Taiwan to Israel are attacking pieces of that problem — but the full-stack humanoid still needs systems integration at industrial scale.

Power and uptime: the silent bottleneck

Humanoid robots are energy hungry: bipedal locomotion, torque-hungry manipulators and onboard compute for perception add up. Solutions that work for wheeled AMRs don’t map directly to bipedal machines. That’s why wireless, on-route charging is getting attention — CaPow’s Genesis platform earned a CE mark and claims it can remove downtime and shrink fleet sizes by about 30% in material-handling trials (installation takes 10–20 minutes per pod). https://www.therobotreport.com/capow-genesis-power-in-motion-tech-receives-ce-mark-european-market/

Perception, autonomy and TRL: software is necessary but not sufficient

CaPow’s CEO Mor Peretz bluntly framed the business case: “We guarantee 100% uptime to 100% of the fleet today,” a bold claim tied to lower-power transfers aimed at AGVs and AMRs. For humanoids, that suggests a hybrid approach: power-in-motion for base-level tasks and fast-swap or tethered provisioning for force-intensive work. Without such strategies, humanoid uptime will remain constrained by battery energy density and workplace acceptance of charging infrastructure.

Beyond energy transfer, designers must rethink envelope and thermal management. More compute (for LLMs and vision stacks) raises heat dissipation needs, and the cost of redundant battery packs or hot-swap systems increases system mass — which then raises power draw in a feedback loop. That loop is why companies pursuing humanoids must partner with power specialists early, or risk building impressive dexterity that can’t be sustained through a typical shift.

Perception, autonomy and TRL: software is necessary but not sufficient

Where humanoids can win first: industrial niche stacks, not generalists

SoftBank’s strategic language about “physical AI” masks an engineering reality: perception and task planning at human scale demand multimodal sensor fusion, deterministic safety behavior, and validated edge compute. YUAN’s ARC AI Platform and a Jetson Orin–based partner stack (Isaac ROS, DeepStream) illustrate the current approach — offload large models to edge accelerators to keep latency low for mission-critical decisions. https://www.therobotreport.com/yuan-unveils-next-gen-ai-robotics-powered-by-nvidia-for-land-sea-air/

But perception is only one axis. Industry uses of autonomy — for example, Energy Robotics’ platform, which has completed more than 1 million inspections and claims 32,000 hazardous-hours saved — show how domain-specific autonomy can reach production TRLs when tasks are narrow and repeatable. https://www.therobotreport.com/energy-robotics-secures-series-a-scale-critical-infrastructure-inspections/ Humanoids, by contrast, must operate across unstructured spaces and social contexts; that raises safety cases, verification burden, and regulatory scrutiny.

Expect a phased TRL trajectory: manipulation primitives and environment observation will mature to TRL 7–8 in constrained industrial cells (proof-of-deployment). Full workplace autonomy — safe, certified operation in populated facilities — will likely sit at TRL 4–6 today and require multi-year testbeds, standardized safety protocols, and certifiable behavior libraries before regulators and customers sign off.

Where humanoids can win first: industrial niche stacks, not generalists

Sources

• The Robot Report — ABB Group to sell ABB Robotics to SoftBank for $5.375B - The Robot Report (2025-10-08)

• The Robot Report — CaPow Genesis power-in-motion tech receives CE Mark for European market - The Robot Report (2025-10-06)

• The Robot Report — YUAN Unveils Next-Gen AI Robotics Powered by NVIDIA for Land, Sea & Air - The Robot Report (2025-10-08)

• The Robot Report — Reframe Systems wants to use robotic microfactories to change how we build homes - The Robot Report (2025-10-07)

• The Robot Report — Energy Robotics secures Series A funding to scale critical infrastructure inspections - The Robot Report (2025-10-07)