Mapping the Physical AI Value Chain Beyond Humanoids

The physical AI value chain includes models, processors, sensors, motion components, robot platforms, integrators and fleet operators. Assess company exposure by the role performed and the revenue earned from it, separating current customer activity from development opportunities. A robot manufacturer is one possible location of value and capital risk within that chain.

Roles and revenue in the physical AI value chain

A useful company map begins with the work performed by the machine. Moving goods, inspecting equipment, assembling parts and supporting surgery require different combinations of software, hardware and human supervision. The components shared across these applications may offer broader commercial exposure than any single platform. However, supplying an enabling technology does not mean all of the supplier’s revenue belongs to physical AI. Economic relevance depends on the actual product, customer and revenue relationship.

Separate the layers of the physical AI system. Models interpret observations and select actions; processors execute workloads; sensors establish the observable state; actuators and mechanical components perform movement. Controllers, safety functions and communications coordinate the machine. Integrators connect it to customer workflows, while operators finance and maintain useful capacity. Some companies occupy several layers. Record each role without counting the same revenue repeatedly when estimating a market or comparing corporate exposure.

Product documentation establishes technical participation before financial materiality. NVIDIA’s Jetson Thor platform is direct evidence of a robotics-oriented compute offering. It does not disclose how much of NVIDIA’s consolidated business is generated by deployed robots. A chip used in both machines and broader computing should therefore have verified enabling exposure and an undisclosed robot-specific revenue share unless filings support something more precise. The same discipline applies to industrial vision, batteries and network equipment.

Ownership, commercial maturity and capital needs

Ownership and channels can matter as much as the component. Rockwell’s completed acquisition of Clearpath Robotics and OTTO Motors in 2023 connected mobile robotics with an established automation business. That transaction evidences strategic integration and access to customer relationships. It does not make every Rockwell sale an AMR sale. A company map should identify the acquired activity, its reporting perimeter and the limits of segment disclosure.

Classify maturity separately from role. A research model, development kit, customer pilot, accepted installation and repeat order represent different stages. A private humanoid developer can have a well-documented prototype but little disclosed recurring revenue. An established motion supplier can earn current cash from conventional automation while preparing components for new embodiments. Both may be relevant, but their valuation sensitivity differs: one depends heavily on future adoption, the other on present industrial demand and incremental product opportunities.

Revenue models explain where capital remains. A component seller finances inventory and qualification but usually transfers the finished asset to the buyer. A systems integrator may carry project working capital until acceptance. A robot-as-a-service provider can retain fleet ownership, maintenance and idle-time risk. Software and data services still require development, support and customer integration. Compare cash conversion, gross profit and capital employed within compatible models instead of assuming that an AI connection creates software-like economics throughout the chain.

Estimating exposure and shared dependencies

An illustrative mapping of a diversified company might identify 20% of sales in a disclosed robotics segment, another 15% in sensors whose robot use is not separated, and 65% elsewhere. The supportable direct share is 20%; the sensor share is an additional area of uncertain exposure, not evidence that 35% of revenue comes from robotics. A range can be useful when the estimation method is explicit. False precision is particularly misleading where customers do not disclose the eventual use of components bought through distributors.

The map should also retain dependencies. A robot platform may depend on a single reducer configuration, an imported processor, proprietary training tools and one integrator’s service team. Multiple company names can therefore represent a concentrated underlying risk. Follow customer acceptance, production releases, support capacity and repeat purchases as the evidence develops. The strongest exposure analysis identifies which part of the value chain earns cash when a machine performs useful work, and which part must continue investing before that work becomes reliable and commercially repeatable.