Editor’s note: This article was written by Lucia, a TechNode reporter.
The first time many people saw a humanoid robot, it was probably doing something impressive: dancing, flipping, or performing a carefully rehearsed routine. The industry’s biggest challenge now lies in turning those impressive demonstrations into robots that can perform useful tasks in everyday environments.
That is the challenge facing China’s rapidly growing humanoid robot industry. As companies move from demonstrations toward commercial deployment, they are exploring how these machines can take on practical roles in workplaces, services and other real-world settings.
Chinese companies are moving faster and at greater scale than U.S. rivals at this stage, Selina Xu, China and AI policy lead in Eric Schmidt’s office, told TechCrunch. But when does the sector get its “ChatGPT moment” — the breakthrough Sam Altman says could be just a few years away?
The Story Behind China’s Dominance
Three forces are giving China an early edge: cost, capital and a large testing ground for commercialization. Its manufacturing base, reinforced by the EV supply chain for batteries, sensors and other components, lets robot makers source hardware locally, iterate faster and lower prices. In the first half of 2026, five Chinese companies accounted for 86% of global humanoid shipments, according to Counterpoint Research.
Funding is another factor. XPeng’s robotics unit raised more than $900 million in August, valuing it at more than $6.3 billion. Galbot raised RMB 2.5 billion ($362 million) in March after a $300 million-plus round in late 2025 valued it at about $3 billion.
China also offers a broad range of potential markets. Humanoids are being deployed or tested in automotive manufacturing, electronics, logistics, aerospace and energy. But commercialization remains early, with deployments ranging from purchases to preorders and pilots, while large-scale repeat orders remain limited.
The Hard Part of the Robot Boom
China’s humanoid makers still face questions over AI systems and integrated software. Vision-language-action models and world models remain early, while Nvidia leads with an end-to-end robotics software stack, leaving many Chinese startups reliant on its Orin chips even as domestic chipmakers develop alternatives.
The harder test is turning physical capability into sustainable commercial demand. Jiang Han, a senior researcher at the Pangoal Institution, told TechNode that repeat orders and customer payback periods are key signs that robots are solving real problems rather than simply attracting trial use. He said China’s longer-term advantage lies not in low manufacturing costs alone, but in combining supply-chain cost advantages with rapid hardware and algorithm iteration.
Data remains another bottleneck. Unlike large language models, robot developers cannot simply scrape the internet for examples of physical interaction. They are turning to synthetic data, simulation, reinforcement learning and real-world deployments. Harry Mellsop, co-founder of Antioch, a startup building simulation tools for physical AI, has described physical AI as being in its “GPT-2 era,” referring to the OpenAI model that predated ChatGPT, with more data and computing power still needed.
Reliability can break an otherwise impressive demo. Speaking at BEYOND Expo, Fu Sheng, chairman and CEO of Cheetah Mobile and chairman of service robotics company OrionStar, said robotics’ hardest challenge is “the last 1%”: a 99% success rate still means one failure in every 100 attempts. He argued that commercial robots must prove they can operate reliably and efficiently over long periods before they become useful workers.
Safety and security add another hurdle. A high-profile accident could trigger public backlash as deployment accelerates. Jiang also pointed to supply-chain vulnerabilities and data-security compliance as gaps Chinese robot makers still need to address, particularly in European and U.S. markets.
Useful First, General Later
Yuli Zhao, chief strategy officer at Galbot, a Chinese humanoid robotics startup that focuses on embodied intelligence and commercial deployment, expects demand to emerge first in manufacturing, warehouse logistics and retail, where tasks are repetitive and workflows are clear — conditions he said create real demand and give humanoid robots a better chance to deliver value at scale.
Fu makes a similar case for specialization. Rather than chasing a general-purpose robot, he argued that companies may find more commercial value in well-defined jobs such as agriculture, transport and sorting, then improve reliability, efficiency and deployment around those tasks.
But the longer-term goal remains general intelligence. Jiang said a true “ChatGPT moment” would require a general embodied-intelligence model that can understand natural-language instructions and break complex tasks into steps rather than rely on preprogrammed actions. The key bottleneck, he said, is handling long-tail situations in unstructured environments, where robots can still fail when something unfamiliar happens.
China may be well positioned to run that cycle quickly. Zhao described its advantage as “speed-to-scale,” with R&D, supply chains, manufacturing, integration and deployment operating in a tight loop — moving robots from prototypes into real-world use faster and feeding operational data into the next round of development.