China’s Humanoid Robots Flood The Market As Shipments Explode

Chinese vendors now ship the vast majority of the world’s humanoid robots. Numbers tell the story with striking clarity. Global shipments hit roughly 19,100 units in the first half of 2026. That marks a 272% jump from the same period a year earlier. Chinese makers claimed more than 97% of them.

Numbers Don’t Lie

AGIBOT alone moved about 8,400 units in those six months. It grabbed a 44% share of the entire global total. Unitree followed with around 5,900. Together the pair accounted for three-quarters of all shipments worldwide. Data from Smart Analytics Global laid out the figures in a fresh report cited by Bloomberg.

The pace has accelerated fast. In full-year 2025 worldwide deliveries reached only about 13,000 to 14,000 units. Chinese firms already held 80% to 90% then. AGIBOT and Unitree each crossed 5,000 that year according to multiple trackers including Omdia. U.S. players such as Figure AI and Tesla managed just a few hundred apiece. Fortune reported the gap plainly.

Analysts now project 60,000 units for all of 2026. Some forecasts from Morgan Stanley have been revised upward twice this year to around 50,000 for China alone. By 2030 the world could see half a million humanoids shipped annually. Revenue estimates for this year hover near $1.6 billion. The long-term prize looks enormous. Yet the present reality mixes rapid volume growth with persistent questions about real capability.

Production capacity tells part of the tale. Unitree and AGIBOT each eye annual output well above 10,000 units now. Chinese suppliers dominate motors, actuators, batteries and sensors. Local costs run 20% lower than foreign equivalents in many cases. Government policy backs the push. Beijing’s plans target a complete domestic supply chain by 2027. The country faces labor shortages and an aging population. Factories, warehouses and state-owned enterprises absorb many of the machines.

But here’s the catch. Many units still serve narrow or demonstration roles. Fortune described the bulk as performative rather than fully functional. They dance at events. They act as museum guides or security guards in controlled settings. They sort parcels in structured postal centers. Real-world messiness trips them up. Dexterity remains limited. Safety concerns linger in unpredictable spaces. Training data stays insufficient for broad generalization.

EngineAI sells basic models for $26,600. Some Chinese humanoids go for under $6,000 in limited runs. Matrix Robotics prices its MATRIX-3 near $99,000 and reports about 1,000 orders from hospitality and retail. Allan Zhang, a founder and ex-Tesla engineer, sees demand in coffee chains and hotels. Yet executives admit the industry sits at an early stage. “We need data from wide scenarios,” one told Fortune. Another warned that without stronger market pull, true mass production will stall.

Over 140 Chinese companies now compete in the space. More than 330 distinct models exist. UBTECH, Leju, Fourier Intelligence and others fill out the top ranks. Many ship to universities for research or to domestic firms for pilot programs. Roughly 85% of demand originates inside China. Overseas sales do occur but trail far behind. The home market provides both scale and a forgiving test bed.

Contrast that with the United States. American firms focus on software and artificial intelligence. They talk of general-purpose machines that learn on the job. Tesla’s Optimus, Figure’s robots at BMW plants and Boston Dynamics’ Atlas draw attention. Their shipment counts remain tiny by comparison. Geopolitical tensions add friction. U.S. restrictions on certain imports and technology transfers complicate the picture.

So what happens next? Costs continue to fall with volume. Average prices could drop toward $21,000 by mid-century according to Morgan Stanley projections. Embodied AI models improve. New applications emerge in auto assembly, logistics and elder care. Yet analysts caution against overhyping near-term breakthroughs. “Use cases still so limited,” one expert said in the Fortune report. “There’s a long way to go” for machines that operate reliably outside labs or tightly scripted environments.

Recent updates reinforce the trajectory. Smart Analytics Global data shows the first-half surge and points to sustained Chinese leadership. TechRepublic covered the same figures on August 11, 2026, noting a shift toward more than 70% of shipments heading into industrial and commercial settings. That marks progress from the prior year. Reliability, manipulation skills and safety standards still require major advances. Household robots for everyday chores appear years away.

Investors pour money in. Unitree pursues a public listing at a multibillion-dollar valuation. Startups raise hundreds of millions. The supply chain matures quickly inside China. Data collection accelerates through real deployments even if many remain limited in scope. Policy coordinates efforts across ministries. The 2025 Humanoid Robot Action Plan sets deployment targets for later this decade.

Challenges abound all the same. Training robust AI demands diverse, high-quality data from thousands of real tasks. Current fleets operate mostly in structured spaces. Fragility raises maintenance costs. Battery life and payload limits constrain usefulness. Western observers note that hardware leadership does not yet equal software or intelligence superiority. The race splits along those lines. China builds the bodies at scale. Others race to supply the brains.

Industry watchers track intervention rates, uptime and total cost of ownership in early commercial pilots. Success in warehouses or factories could unlock bigger orders. Failure to move beyond demonstrations risks a glut of underutilized machines. So far the volume story favors China overwhelmingly. Whether that translates into economic impact or sustained competitive advantage remains the open question.

One thing looks clear. The era of experimental humanoids has given way to volume manufacturing. Chinese vendors drive that transition today. Their output dwarfs everyone else’s. The test ahead lies in making those robots genuinely productive. Not just numerous.