Shanghai-based robot maker Agibot, backed by Chinese tech giants Tencent, BYD, and Baidu, said on December 7 it has reached a milestone of producing 5,000 humanoid robots at its flagship factory since it was founded in 2023, a figure that places the startup among the world’s biggest producers of such products by shipments.
In just a few short years, Agibot has emerged as a leading humanoid robot builder under its 32-year-old co-founder and president, Peng Zhihui, a nationally renowned engineer who previously worked at major Chinese tech companies, including Huawei and Oppo.
Goldman Sachs and BofA Global Research estimate that humanoid robot shipments will reach about 18,000–20,000 units in 2025. The figure for 2024 was only about 3,000 units, meaning any manufacturer capable of producing even a few thousand units is already making a meaningful impact on the market.
Qiu Heng, Agibot’s CMO, told Nikkei Asia that the robot maker has a vision of building humanoid robots with “general intelligence,” rather than only performing specific tasks.
Qiu said they are progressing step by step with vision-language-latent-action (ViLLA) framework models developed specifically for training different types of humanoid robots. Training robots requires interaction not only through language, but also through motion, vision, and other senses. Its robots can now handle simpler tasks such as dancing, performing tai chi and giving introductions at exhibitions, as well as conducting patrols or inspections, collecting data, and carrying out certain specialized tasks in a factory.
“Performing dances is more of an entry-level capability, while handling factory tasks is considered more advanced,” Qiu said. “Our long-term vision is to build robots that are not like fixed robot arms limited to specific tasks. Instead, we hope they can be more humanlike, able to cook one day, work in a factory the next, drive a car another day, and continue learning new skills over time.”
Qiu said his company believes humanoid robots will enter households and start to become part of consumers’ daily lives in “three to five years.” All of Agibot’s robots can receive software updates and become more intelligent and skilled, he added.
Beijing-registered Galbot, founded by robotics scholar Wang He and backed by food delivery group Meituan and top battery maker Contemporary Amperex Technology (CATL), is taking a slightly different approach. Instead of setting broad ambitions for robots that can do everything, Galbot is focusing on three targeted training models, but prioritizing retail and industrial manufacturing applications.
“Currently, the most mature applications for our humanoid robots are in the retail sector. We have deployments in several stores in Beijing and will deploy massively in retail sectors in many major cities across China,” Xiao Minpeng, an overseas marketing representative with Galbot, told Nikkei Asia. “Within that specific domain and scenarios, we have our own trained models for retail shelf operations.”
He said Galbot is working more on wheeled humanoid robots rather than four-footed patrol bots or dancing humanoid models. “Our approach is more down-to-earth, to master skills rather than rolling out so many different types of solutions all at once,” he said.
Galbot and Agibot are among Nvidia’s early strategic partners, using the US chipmaker’s platforms to develop and deploy humanoid applications. Other leading Chinese players, including Unitree and UBTech, are also collaborating closely with Nvidia. Key robot players in the US include Tesla’s Optimus, Agility Robotics and Boston Dynamics, majority owned by Hyundai Motor Group, while major Asian players like Xiaomi, Xpeng, and Samsung are also making their own bets on humanoid robotics.
Major industrial robotic solutions providers in Europe and Japan, such as ABB, Fanuc and Kawasaki Heavy, dominate autonomous robotic arms for factory use but are far less active in developing broadly intelligent, general-purpose humanoid robots, according to industry executives and analysts.
Hong Kong-listed Dobot, one of China’s leading industrial robotics solution providers, has also invested in developing humanoid robot models, though the company acknowledges the nascent industry is facing several bottlenecks.
For example, humans are more than capable of handling tasks such as working at specific factory stations, “and the cost of building and deploying humanoid robots still cannot compete with the cost of human labor,” a Dobot representative told Nikkei.
Still, many analysts say the market for humanoid robots is enormous and that embedding artificial intelligence into physical systems represents the next major frontier of technological growth. According to Morgan Stanley, global cumulative humanoid adoption could reach one billion units by 2050, with adoption growing slowly through the mid-2030s and accelerating rapidly after 2035 as hardware, software, and AI models mature enough to enable truly general-purpose humanoid robots.
The industry also faces a core challenge known as Moravec’s paradox, which refers to the observation that while computers can relatively easily achieve adult-level performance on tasks such as intelligence tests, it remains extremely difficult to give robots the perceptual and motor skills of a one-year-old child.
Agibot’s Qiu said certain components, such as a “dexterous hand” capable of performing advanced skills and various joints and actuators on the robot’s body, will be very important for the future.
One key hurdle to adoption is heat dissipation issues for the “brain,” “joints,” and battery of the humanoid robot form, according to Sharon Shih, an analyst with Morgan Stanley.
While optimism about physical AI is running high—buoyed by statements from figures such as SoftBank founder Masayoshi Son, who has called it the next big wave in tech—others are more cautious. Former Intel CEO Pat Gelsinger said humanoid robots still have many technological challenges to overcome and that progress could go through up-and-down cycles.
“The biggest challenge for humanoid robots today is having humanoids safe in the presence of humans,” Gelsinger told Nikkei Asia. “Consumer humanoid robots are not going to happen [any time soon].”
