Global humanoid robot shipments reached roughly 13,000–13,300 units in 2025, up from about 2,300 in 2024. Chinese makers dominated the volume: AgiBot shipped about 5,168 units, Unitree around 4,200–5,500, and UBTech about 1,000, while US firms Figure AI, Agility Robotics, and Tesla each shipped roughly 150.
China's lead is attributed to government policy support, public investment, mature manufacturing supply chains, and advances in AI hardware and software. However, multiple reports caution that many of these robots are still more impressive in demos and dance routines than in reliable industrial work, so shipment numbers may flatter the real state of the technology.
Omdia counted about 13,300 humanoid robots shipped globally in 2025, up from roughly 2,300 in 2024, with Chinese firms such as AgiBot and Unitree dominating volume.
Figure AI is indeed using its Helix vision-language-action model for general-purpose humanoid intelligence. But the claim that the race “isn't about demos anymore” is too strong: recent reporting says many humanoid robots remain better at demos and dance routines than reliable industrial work.
Boston Dynamics has shown Atlas walking, running, crawling, recovering from falls, and performing handstands, using reinforcement learning developed with the Robotics and AI Institute.
Figure AI's BotQ facility is designed for about 12,000 robots per year, and 1X Technologies is scaling production. Tesla has ramped Optimus output, with reporting of several hundred per week by August 2026, but the “Optimus Gen 3” label is not established in the cited reporting. Link to a fan-run browser version of Need for Speed Underground 2; no factual claim to check.
The Guardian reported that Apple's autocorrect has an internal “grey list” of words and phrases it avoids helping users spell.
CThousands of humanoid robots shipped in 2025. Many of them from China.
MThirteen thousand sounds impressive until you ask what those robots actually do.
CKrosh posted the side-by-side on X— an American humanoid, then a Chinese one. His point: the gap isn't movement, it's numbers.
MHe's right. Omdia counted about thirteen thousand three hundred units in 2025, up from two thousand three hundred in 2024.
CAgiBot alone shipped more than five thousand. That's several times the entire US output.
MFigure, Agility, and Tesla each shipped roughly one hundred fifty. It's a manufacturing blowout.
CRest of World reported China's share at ninety-seven percent of global shipments. That should wake people up.
MIt's a manufacturing story, not just a tech story.
CBut shipping a robot is not the same as doing the job. Elina Notes on X made a related claim.
MWhat's the claim?
CShe says the race isn't about demos anymore. It's AI intelligence, manufacturing scale, reliability.
MThat's a nice speech. But demos are still very much the story.
CAIOSTIK posted on X— Boston Dynamics' Atlas walking, running, crawling, recovering, doing a handstand.
MExactly. A handstand is a demo. Chinese factories are shipping thousands.
CA handstand with reinforcement learning is not just a party trick. It's balance, recovery, real-time control.
MMaybe. But Reuters went inside Chinese firms and found many bots too slow and error-prone for factory floors.
CSo the shipment numbers flatter them. That's why I'd still bet on the American brain.
MYou can't export a brain. If China owns the supply chain, they iterate on software while we polish demos.
CNo, wait— that's backwards. Software is the moat. A warehouse full of hardware without intelligence is just statues.
MStatues that already ship. Omdia's analyst Lian Jye Su says China's edge is policy, investment, supply chains.
CAnd Elon said China is very good at AI and manufacturing— Tesla's toughest competition. He's not dismissing it.
MLian Jye Su also says Western firms can still compete on AI, software, and autonomy. In 2025, Tesla shipped far fewer units than Chinese firms like Unitree.
CFine, scale is real. But what do we want— a hundred robots that actually work, or ten thousand that dance?
MI'd like one that works AND gets built. The US is not building at scale yet. That's the actual story.
CFigure's BotQ plant is designed for twelve thousand robots a year. American firms are trying to catch up.
MA designed capacity is not a shipped robot. That's a press release with a clean floor.
CFair. But Figure does have the Helix model for general-purpose intelligence. That's software, not a press release.
MRight. Figure bets on the brain. Boston Dynamics shows the body. China ships the bodies.
CAnd for now, the bodies win. But this race is just getting started.
MIf the US doesn't fix manufacturing, it's over before it starts. That's what worries me.
CYou usually worry about hype cycles. Now you're worried we're not building enough.
MDifferent problem, same result— losing to China because we thought clever was enough.
CSo what's the fix? Industrial policy? Subsidies? Because that's exactly where China beat us.
MNo, private capital is already moving into Figure and 1X Technologies. But they need customers, not just demos.
CCustomers need reliability. Reuters said many bots aren't ready for industrial work yet.
MExactly. So the number to watch isn't shipments. It's hours worked on a real line.
CI'll give you that. The next twelve months: who logs real factory hours, not handstands.
MAnd if China fixes the software, those thirteen thousand robots start working. Then it's a problem.
CThat is the real race. Not jumping higher— doing the job.