Size up the humanoid robotics race

The humanoid robotics industry is real, growing, and substantially overhyped—simultaneously. Approximately 13,000–19,000 humanoid robots shipped globally in 2025–H1 2026, but roughly 85–90% were Chinese-made, predominantly delivered to universities, research labs, and state-owned enterprises rather than productive factory floors. Among Western companies, only Agility Robotics has demonstrated the broadest verifiable record of repeated, paid commercial deployment with named customers. Figure AI completed a meaningful industrial pilot at BMW but carries a $39 billion valuation on effectively zero audited revenue. Boston Dynamics Atlas is entering production but all 2026 output is committed internally. Tesla Optimus remains an internal R&D program with serially missed targets and no external sales. UBTECH and AgiBot lead in production volume but most use cases are not industrial-productivity deployments. Physical Intelligence (π) is a pure-software play—well-funded, technically credible, pre-revenue, and not yet deployed at scale. Sanctuary AI has pivoted from humanoid hardware to selling an AI software brain for existing industrial robots, an honest strategic adjustment rather than a simple failure. The field as a whole is in what Bessemer Venture Partners calls a "GPT-2.5 moment"—past the pure demo phase, short of reliable economic deployment.