Humanoid robots close to 'ChatGPT moment'

The humanoid robot industry could hit its "ChatGPT moment" by the end of 2028, based on current data accumulation rates and model convergence trends, said Wang He, founder and CTO of Beijing-based Galbot Co Ltd.

The "ChatGPT moment" has become a popular metaphor in tech and investment circles to describe a sudden breakthrough that catapults a new technology from niche circles into the public consciousness, allowing ordinary people to viscerally feel its practical value.

Wang said a powerful pre-trained model combined with an efficient post-training paradigm is the key to unlocking this milestone, and embodied artificial intelligence foundation models trained on massive, comprehensive datasets could achieve a 70 to 80 percent success rate directly for tasks they have never been specifically trained for.

Galbot was among the invited enterprises to a private enterprise symposium hosted by Zheng Shanjie, head of the National Development and Reform Commission, in Beijing on Monday. The event focused on analyzing first-half economic performance and gathering suggestions for second-half economic work.

The policy attention comes as Galbot's technology is capturing the public imagination in tangible ways. At the 2026 World Artificial Intelligence Conference, which concluded in Shanghai on Monday, crowds three or four layers deep gathered around Galbot's exhibition area. They were watching Galbot G1, one of the company's robots, make breakfast, toast bread and pour drinks with the smooth coordination of a seasoned short-order cook.

But what truly captivated the audience was not the actions themselves — it was what happened next. Spectators repeatedly moved the bread aside, shifted the kettle, rearranged cups and plates, and covered the cup lids with their hands, creating obstacle after obstacle. Yet the robot never wavered. It relocated its targets in milliseconds, reconstructed its path on the fly and completed every step without interruption.

Wang outlined a roadmap for general-purpose embodied intelligence foundation models. He said that building a unified "brain and cerebellum" — the embodied intelligence foundation model — to achieve a deep understanding of human movement and control mechanisms represents a critical technological pathway for pushing humanoid robots toward the "tipping point" of artificial general intelligence.

Just a day before the WAIC opened, Vice-Minister of Industry and Information Technology Ke Jixin revealed at the annual plenary session of his ministry's humanoid robot and embodied intelligence standardization technical committee in Shaoxing, Zhejiang province, that China's humanoid robot output reached about 20,000 units last year, surpassed 40,000 units in the first half of this year, and is expected to exceed 100,000 units this year.

This fivefold jump sends a clear signal: humanoid robots are accelerating from technological exploration toward standardized, large-scale development. But quantity expansion does not equal industrial maturity.

Xie Shaofeng, chairman of the ministry's technical committee and chairman of the OpenAtom Foundation, warned that the industry is currently "blossoming yet fighting separately". Behind the apparent prosperity are three major challenges: highly divergent technological pathways, a critical lack of product standards and mounting pressure from sustainable development requirements.

Ke said the next two to three years could be a critical turning point for the humanoid robot sector. Output growth is merely the surface — the deeper shift is that the industry is moving from single-unit prototypes and stage demonstrations toward batch deliveries, which dramatically raises the coordination requirements across research and development, production, testing, application and safety governance.

At the prototype stage, companies can solve interface, algorithm and hardware adaptation issues through customized solutions. But once production reaches the scale of tens of thousands of units, any interface incompatibility, inconsistent evaluation methods, or unclear safety boundaries will be magnified into cost, efficiency and market trust problems, experts said.

To a certain extent, the core of scaling is not just "expanding production capacity" but also enabling components from different companies to work together, allowing product performance to be compared and contrasted, thus giving users the confidence to buy and use products and providing a basis for after-sales maintenance, they added.

Ke called for leveraging the foundational, guiding and supportive role of standards, as well as accelerating the development of urgently needed standards for humanoid robot intelligence classification, data management and safety specifications.

Jiang Lei, deputy director of the technical committee, said hardware requires standards and models need "firewalls", adding that the committee expects to roll out the first batch of standards by the end of this year.