China industrial robots move from gala stunts to gigafactories

From viral performance to unmanned services

At this year’s China International Fair for Trade in Services (CIFTIS) in Beijing, the humanoid robot Galbot G1 moved beyond its Spring Festival Gala fame for folding laundry. It took on a more commercial role staffing an unmanned convenience pod. According to Xinhua’s report on the event, similar pods already operate at sites including Beijing’s Summer Palace, where each is expected to serve more than 1,000 visitors a day. This pushes service robots into high-footfall, revenue-generating locations rather than confined demos.

Meanwhile, the model is starting to cross borders. Industry coverage of Galbot’s expansion in Hong Kong shows that the first overseas convenience stores operated with Galbot humanoid robots held their opening around the end of August. Outlets are scheduled to begin serving customers from 1 September. This marks an early export of China’s robot-run retail concept and gives investors a live test of its unit economics outside the mainland.

On the CIFTIS floor, robots were visible across everyday tasks — from an unmanned post office booth where a robot handled stamping and mailing, to pancake and coffee robots that manage food service with consistent quality. For China industrial robots makers and operators, these deployments offer scalable use cases in retail, logistics and food — all price-sensitive but high-volume sectors.

A manufacturing base that already runs at scale

The industrial ambition runs deeper. China already operates more than two million robots in its factories, according to data cited by the BBC from the International Federation of Robotics. The country produced 538,000 industrial robots in the first half of 2026, up 28 percent year on year. That base gives embodied AI developers an existing infrastructure footprint and buyer network as they push humanoid and service robots beyond show business.

Export momentum is equally striking. China became a net exporter of industrial robots in 2025. In the first half of 2026, exports reached 6.29 billion yuan, up 18.6 percent year on year and sold to 141 countries, according to State Council briefing data. Scale and price competitiveness are opening markets that domestic peers in Japan, Germany and South Korea have held for decades.

What does the training-data shortage mean for investors?

However, the bottleneck is clear. In the first half of this year, China’s domestic shipments of humanoid robots reached 23,000 units. Fewer than 5 percent are working stably on real production lines, according to Zhu Kai, head of the Shijingshan Embodied Intelligence Training Center in Beijing. Zhu links this gap directly to data: high-quality multimodal real-robot data for industrial training does not reach even 1 percent of actual demand.

As a result, Shijingshan’s centre operates as a ‘vocational school’ for robots. Nearly 270 robots work at full capacity on chores, caregiving drills and factory tasks such as placing circuit boards on production lines. They generate close to 20 million high-quality training data entries each year, per the Xinhua Economic Watch report. These datasets capture human-like movement, tactile feedback and error conditions, then feed into large models that guide robots on real-world production lines.

This school is backed by a wider data infrastructure push. The Shijingshan district has registered what Xinhua describes as China’s first 3D high-quality dataset for large-model training, opened four data service stations and listed more than 200 data products. Beijing targets 100 high-quality industrial datasets by 2028. For investors, this positions embodied-intelligence data as a tradable asset class. Platforms aiming to provide international services for embodied-intelligence data were launched at CIFTIS in partnership with China Telecom‘s Beijing branch.

New occupations signal industrial policy, not just hype

The labour market is being adapted to support this new infrastructure. On 9 September, China officially recognised embodied-intelligence robot application technicians as a new occupation, along with 10 other new roles. The recognition came in a notice from the Ministry of Human Resources and Social Security. Within this occupation, two new trades stand out: robot data collectors and robot trainers.

These roles cover tasks from designing data capture schemes and recording multimodal interaction data to tuning models and monitoring field performance. An analyst view from Beijing captures the shift crisply: ‘Once data work becomes a formal occupation, embodied AI moves from experiment to industrial policy.’

The services dimension is already global. Anyverse Dynamics, a Beijing embodied-AI start-up at CIFTIS, signed a 500-million-yuan deal — about US$74.7 million — with green-tech group Envision. The agreement sends its K15 semi-humanoids to Envision’s battery gigafactory in France for autonomous loading and unloading work. The robotics hardware is only part of this revenue. Ongoing training, monitoring and software updates form a recurring services stream that aligns with China’s broader push to expand its services trade.

As embodied-intelligence platforms and technician occupations mature, investors should watch three things next: how fast deployment rates close the training-data gap, how embodied AI data platforms monetise cross-border services, and how China’s new robot vocational schools reposition both labour and capital in the country’s next phase of industrial automation.

Quick answers