Humanoid robots drive near-term chip demand before mass production

The humanoid robotics market has attracted investor attention largely on visible hardware—motors, actuators, sensors, and batteries—but Barclays' analysis of over 100 companies across the value chain reveals that the real near-term constraint is intelligence. Building robots that can perceive and act in unpredictable environments requires a three-layer computing stack: data-centre infrastructure for simulation and training, foundation models for perception-to-action translation, and on-robot processors for real-time local decision-making.

Simulation plays a critical role because developers lack the internet-scale datasets available for training language models. Digital environments can teach robots locomotion and object handling before real-world deployment, but physical testing remains necessary to validate what simulations cannot accurately reproduce. This means demand for compute will likely scale before robot production itself accelerates as developers train models and create digital twins. The timeline matters: while the humanoid market is currently $2 billion to $3 billion with forecasts reaching $10 billion to $25 billion by 2030, large-scale economic deployment may be closer to 2035, constrained by unresolved safety, reliability, and autonomy challenges.

For semiconductor suppliers, this creates an early revenue window. Nvidia, already offering an integrated robotics stack (Omniverse for simulation, Isaac Sim for training, GR00T foundation models, and Jetson processors), stands to capture demand for compute infrastructure before hardware production constraints ease. As manufacturing scales toward 2035, actuators—currently 30% to 50% of component cost—will become more critical than onboard compute (now 10% to 15%), but low current volumes limit standardisation and supplier investment.