Yet the biggest near-term constraint is intelligence, as humanoids must perceive their surroundings, reason and adjust their behaviour in unpredictable environments. Demand for compute may scale before robot production as developers train models and create digital twins. Nvidia already provides an integrated robotics stack spanning Omniverse for digital environments, Isaac Sim for training, GR00T foundation models and Jetson processors for on-robot computing. Actuators account for an estimated 30% to 50% of a humanoid's component cost, compared with 10% to 15% for onboard compute, but low robot volumes currently limit standardisation and supplier investment. Related articlesHumanoids are as much a compute story as a robotics one5 reasons why Jefferies thinks Meta's pullback is a buying opportunityWolfe Research outlines eight risks that could spark stock declines in 2026