China's robotics industry is on the brink of a tectonic shift. ACE Robotics Chairman Wang Xiaogang predicts that by the end of 2027, we will witness the "ChatGPT moment" for embodied artificial intelligence—that pivotal stage when the technology transforms from a laboratory curiosity into a mass-market commercial product. In my assessment, this is an extremely ambitious but quite realistic timeline, given the current pace of development.
Two key factors will drive this breakthrough: fundamentally more advanced robot control models and the exponential expansion of datasets on real physical interaction. However, as Wang emphasizes, full-scale commercial deployment of the technology will take another four to five years after that point.
The Problem Is Not the "Hardware" but the "Brains"
Modern humanoids already demonstrate impressive physical capabilities—they run, dance, and perform highly specialized tasks. But their main weakness is the inability to adapt to unfamiliar environments and a wide range of actions. ACE Robotics rightly points out that the limitation is not the hardware but the AI models. Embodied intelligence systems must perceive their surroundings in real time, interpret instructions, predict consequences, and make decisions about movements.
To address this challenge, the startup is developing the Kairos model, which combines perception, multimodal understanding, physical environment simulation, and behavior planning. As early as July, the company announced first place in four tests for embodied AI. Notably, the open-source Kairos-4B version, despite its modest 4 billion parameters, outperformed giants like Nvidia Cosmos 3 and Ant Group's Lingbot in public benchmarks. This confirms that efficient architecture can matter more than raw computing power.
Data Scarcity—The Industry's Main Bottleneck
The critical problem remains the lack of data on real physical interaction. The entire industry has collected only about 100,000 hours of such recordings in recent years—catastrophically insufficient for training foundation models. The traditional approach of teleoperation in exoskeletons is extremely slow and expensive.
ACE Robotics offers an elegant solution: the company equips ordinary workers on real production lines with lightweight sensors and records their natural actions. Over two years, the startup expects to accumulate tens of millions of hours of data—creating a colossal competitive advantage.
Expansion Plans and a Strategy to Reduce Dependence
The company is already testing its models in unmanned retail, hotel services, and logistics, using humanoids from Unitree, AgiBot, and Fourier. Ambitious plans include deploying the technology in at least 1,000 stores within a year and expanding to 10,000 by the end of the second year. Also notable is the strategy of diversifying computing power: alongside Nvidia accelerators, Chinese chips from Rhino Tech and Digua Robotics are being utilized—this reduces costs and lessens dependence on Western suppliers.
Since its founding in July 2025, ACE Robotics has raised over $100 million from Ant Group and SenseTime and is preparing for an IPO. Against the backdrop of Morgan Stanley's revised forecast (growth in Chinese humanoid shipments in 2026 from 28,000 to 50,000 units), the stakes in this race are extremely high.
My conclusion: ACE Robotics demonstrates a rare understanding that in robotics, the winner is not the one with better "hardware," but the one who provides AI with quality data about the real world. If data collection rates hold, Wang's forecast may even prove conservative.