Ninety-seven percent

China shipped 18,500 of the world's 19,100 humanoids in the first half. Volume leadership is not capability leadership — but it is how capability leadership eventually gets decided.

Shipments tripled to roughly 19,100 units, AgiBot passed Unitree at 44% share, and industrial and commercial uses are now over 70%.

The second number is the important one

Tripling shipments could mean anything — research procurement, demonstrations, subsidised placements. Industrial and commercial applications passing 70% means the category has moved from budget line "innovation" to budget line "capital equipment". Those get renewed on results.

The sceptical case, stated properly

Shipping units proves supply chain, cost and assembly throughput are solved. It proves comparatively little about whether the deployed robots do valuable work reliably, and a robot delivered is not a robot still running in a year. That case is legitimate and currently unfalsified.

Manufacturing leadership and capability leadership can stay apart for a long time. They do not stay apart forever.

Two bets, both funded

Boston Dynamics sold out its 2026 electric Atlas run — to Hyundai's robotics research centre and Google DeepMind. Not deployments. Platforms for developing the technology.

So there are two theories being funded simultaneously. One says humanoid value comes from fleet learning, and accrues to whoever has the most robots generating the most operating data. The other says it comes from per-unit capability, and accrues to whoever builds the best one.

Volume favours the first. Research platforms favour the second. Nobody has to be wrong yet — but the fleet gets a new half-year of data every six months, and the research platform gets a new model.

Semafor — AgiBot passes Unitree as China's top humanoid exporter → · TechRepublic — Humanoid Robot Shipments Surge 272% as Chinese Vendors Capture 97% Share → · SCMP — AgiBot overtakes Unitree as top global humanoid robot vendor →