China’s humanoid robots are getting very good at putting on a show. They can sprint, dance, box, play football and perform feats that would have sounded improbable only a few years ago. But behind the viral videos and increasingly eye-catching demonstrations lies a far more consequential experiment: China is trying to figure out whether these machines can move from spectacle to something the economy can actually use.
That question is becoming harder to dismiss as a futuristic one. What China is building is no longer confined to isolated laboratory prototypes or choreographed tech demos; the scale of participation, investment and experimentation suggests an industry being pushed rapidly towards the mainstream.
For instance, at the second World Humanoid Robot Games in Beijing, over 2,000 robots from 666 teams across 16 countries competed in 51 events in August. The field was overwhelmingly Chinese: 641 Chinese teams brought 1,975 robots, representing 157 companies and 200 universities and research institutions.
The competition also went beyond sports. Robots were tested in emergency response, hotel services, picking up small objects and tightening screws. That is closer to what Chinese policymakers and companies ultimately want: machines that can move from demonstrations into factories, warehouses, pharmacies and other workplaces.
China’s humanoid-robot industry is expanding rapidly. The country now has more than 150 humanoid-robot companies, according to China’s National Development and Reform Commission, Reuters reported. National, provincial and local governments spent at least $230 million on humanoids and related equipment in the first half of 2026, compared with $62 million a year earlier, reported Reuters.
The ambition is therefore much larger than building robots that can win races. China is trying to build the manufacturing, software, training-data and deployment ecosystem needed to create a commercially viable humanoid workforce.
Why did China’s robots attract so much attention?
The biggest headline from the Beijing Games came from Tiangong Ultra. The humanoid, developed by the Beijing Humanoid Robot Innovation Center, completed the 100 metres in 8.86 seconds, reported Reuters. That was faster than Bolt’s human world record of 9.58 seconds set in 2009.
The robot later improved its reported time to 8.64 seconds, according to Chinese media outlet People’s Daily. Its reported 400-metre record stood at 38.15 seconds, while its standing high-jump record reached 3.40 metres.
The improvement over a year was striking. Tiangong had won the 100-metre event at the 2025 Games in 21.50 seconds. One year later, its reported time had fallen below nine seconds.
Other robots also demonstrated rapid progress. Honor’s Lightning recorded a 9.47-second 100-metre run during the Games and had clocked 9.32 seconds in a preparatory test, Reuters reported. It also won the humanoid half-marathon in 50 minutes and 26 seconds.
From football and boxing to factories and pharmacies
China is increasingly testing humanoids in environments that resemble real work. AGIBOT, for example, has focused on production and deployment. Tech market research firm TrendForce earlier stated that the company produced its 10,000th general-purpose embodied robot, the Expedition A3, in March.
The company subsequently ran a six-day factory trial at Longcheer Technology’s plant in Nanchang. AGIBOT said its robots completed more than 64 hours of operation and 64,828 production-line tasks, with a reported 99.99% task success rate. Those figures are company-reported and do not establish that humanoids can replace factory workers broadly.
UBTECH is taking another approach. At a UBTECH-backed training centre in Liuzhou, southern China, Reuters found more than 100 humanoid robots being trained for tasks including sorting crates, packaging noodles and making coffee.
But training remains difficult. A novice trainer could sometimes produce one usable robot movement only once in about 300 attempts, reported Reuters. An experienced trainer could produce one in roughly 50 attempts. The centre received an $18 million tender from the Guangxi regional government to supply humanoids and related hardware, with the objective of collecting training data for embodied AI.
Some robots have already reached limited commercial applications.
Galbot has deployed humanoids in pharmacies across more than two dozen Chinese cities. At one Beijing pharmacy, its robot picks medicines from shelves after receiving digital orders. The company said the robot could pick and prepare an item in about a minute, with a success rate above 95%. Most failures involved difficulty grasping an item rather than selecting the wrong product.
Why is China spending so heavily on humanoids?
The robotics push is being supported by government spending, industrial policy and an unusually large manufacturing ecosystem.
China’s State Grid announced plans in April to spend $1 billion on AI-enabled humanoids, dual-arm robots and robot dogs for grid maintenance and substation inspections. Shenzhen is targeting a $15 billion embodied-intelligence robotics cluster with more than 1,200 companies by next year, according to Reuters report.
The strategy resembles China’s earlier push into electric vehicles and solar panels: encourage investment, bring in large numbers of manufacturers, drive competition and reduce costs.
But humanoids are not electric vehicles. EVs were already a mature commercial product when China’s manufacturing expansion accelerated. Humanoids are still struggling with reliability, dexterity, intelligence and cost. That raises the possibility that production capacity could grow faster than actual demand.
Kevin Xu, founder of Interconnected Capital, described China’s humanoid sector as a potential “bubble”, saying that consolidation was inevitable, as reported by Reuters.
Prices are already falling. Morgan Stanley expects the average price of a humanoid robot to decline 15% in 2026. Leju Robotics’ full-size Kuavo humanoid became 26% cheaper in 2025, falling to about $46,000, Reuters reported.
For buyers, however, falling prices could eventually make the technology more attractive.
China’s biggest advantage may be data, not just hardware
The hardest part of building a useful humanoid may not be manufacturing the body. It may be teaching the machine how to operate in the physical world.
Humanoids need what companies call vision-language-action models. A robot has to see an object, understand what it is, decide what to do, determine how to grip it and then execute the movement accurately.
That is fundamentally different from a chatbot generating text.
The industry currently has roughly 500,000 hours of high-quality training data, while one estimate cited by Reuters puts the requirement for true physical intelligence at around 100 million hours.
Robert Wu, a Shanghai-based technology investor, told Reuters that companies need a large installed base of robots because those machines can collect data from the physical world.
This may explain why Chinese companies are pushing production even before every robot has a clear commercial use.
The biggest problem: Robots can demonstrate skills, but can they work?
This is where the excitement around China’s humanoids needs to be tempered. A robot running 100 metres in 8.64 seconds demonstrates extraordinary physical performance. It does not prove that the machine can work an eight-hour shift.
Real workplaces are unpredictable. Objects arrive at different angles. Lighting changes. Workers move around the machine. Components can be misplaced. Surfaces can become slippery.
Humans adapt to such changes almost automatically. Humanoids still struggle.
China’s humanoids remained too slow and error-prone for most industrial work, reported Reuters. Robots at the Liuzhou training centre completed simple tasks at only about 20% of human speed or output.
Tang Wenbin, co-founder and CEO of AI robotics company Yuanli Lingji, said, “A lot of what we see is dancing disguised as working,” as reported by Reuters.
Conventional industrial robots also remain more efficient for many factory tasks.
At Xiaomi’s electric-vehicle plant in Beijing, 91% of the assembly process is automated, with more than 700 industrial robots performing tasks such as welding, riveting and adhesive bonding. Humanoids are being tested for selected tasks, but conventional automation still handles most production work.
Geely’s Zeekr brand has similarly tested UBTECH humanoids for sorting and material handling, while saying robotic arms and automated vehicles continue to handle all key tasks on the active assembly line.
The challenge becomes particularly difficult when robots need to manipulate small objects.
Which Chinese robot companies are leading the race?
Unitree has become one of the most visible Chinese humanoid makers internationally. The company’s shares surged nearly sixfold on their Shanghai debut before falling 11% the following day, with its market value reaching about $50 billion at one point, Reuters reported.
CEO Wang Xingxing has described the industry as moving towards a “ChatGPT moment” for embodied intelligence, as reported by Reuters. He wants robots that can enter unfamiliar environments and complete a large share of tasks after receiving simple voice or text instructions.
But even Wang has given a cautious timeline. He said a major software breakthrough could arrive within two to three years in an optimistic scenario, or within five to 10 years at the latest.
AGIBOT is focused more heavily on production and industrial deployment, while UBTECH is building training infrastructure and collecting physical-world data.
Where does India stand in robot race?
India is starting from a much smaller industrial base. International Federation of Robotics data shows that Indian factories installed about 9,100 industrial robots in 2024, compared with 295,000 in China. India’s installations nevertheless increased 7% that year. China accounted for 54% of global industrial-robot installations in 2024, and its operational stock exceeded two million units.
China’s established manufacturing base gives its companies access to motors, sensors, batteries, actuators, automation customers and other components.
India’s physical-AI ecosystem is nevertheless beginning to emerge.
Bengaluru-based Ati Robotics raised $37 million and planned a Michigan manufacturing plant for physical-AI robots, reported Business Standard. Its founder Saurabh Chandra told Business Standard that the company was developing an industrial humanoid capable of moving 25-kg bins without teleoperation.
Addverb has tested physical-AI robots for industrial and defence applications. Gurugram-based Novus High-Tech Robotic has also worked on humanoid and quadruped robots for industrial use.
Indian government has also started work on a national robotics roadmap. The Principal Scientific Adviser’s Technology Advisory Group met in February 2026 to discuss India’s robotics ecosystem and a strategic roadmap.
The discussion on the strategic roadmap identified gaps in actuators, sensors, transducers, chipsets, precision gears and design tools. It also called for stronger testing, certification and standardisation, along with a phased approach to indigenising critical components.
These priorities suggest that India is focusing on building domestic capabilities across the robotics technology stack rather than simply measuring progress by the number of robots produced.
Is China creating a humanoid-robot bubble?
There is a significant financial risk behind the manufacturing push. BofA Global Research estimates that about 20,000 humanoid robots were shipped globally in 2025, with Chinese manufacturers accounting for 95%. China’s Ministry of Industry and Information Technology expects the country to build more than 100,000 humanoids in 2026, while BofA forecasts global annual shipments of 1.2 million units by 2030.
Those numbers suggest a potentially huge market. They also raise a question that will demand grow fast enough to absorb the machines being produced?
Government entities often act as both financiers and customers in China’s humanoid industry, reported Reuters. That can help companies scale, but it can also create demand that does not necessarily reflect sustainable commercial demand.
The US-China technology rivalry adds another layer. Nvidia has made physical AI a major part of its strategy, while Chinese robotics companies remain important customers for US technology.
At the same time, Washington has begun treating humanoid and quadruped robots as a national-security issue, with restrictions on some foreign-made robots.
That could make humanoid robotics another technology battleground between Washington and Beijing.