The chairman of the Chinese startup ACE Robotics, Wang Xiaogang, stated that by the end of 2027, a breakthrough in the field of creating “brains” for human‑like robots could occur, comparable to the impact of ChatGPT on the AI industry. According to him, global models and the collection of environmental data will contribute to reaching this turning point. However, even after that, it will take another 4-5 years for widespread commercial adoption of embodied intelligence models in various sectors.

Unlike large language models, physical AI simulation systems are designed to help robots understand and navigate real-world environments in real time. The founder of Unitree, Wang Xingxing, also predicted that a significant breakthrough in robotics could occur in 2-3 years. At the same time, industry leaders acknowledge that obtaining high-quality data for training models in real-world conditions remains the main bottleneck.

ACE Robotics, founded in July 2025 with the support of Ant Group and SenseTime, raised more than $100 million in the first half of 2026. The company intends to go public via an IPO as soon as Chinese listing rules allow it, as they require a minimum of 3 years of financial history. ACE’s open model, Kairos‑4B, ranks first in the world according to publicly available benchmarks, outperforming Nvidia Cosmos 3 and Ant Group Lingbot, despite having far fewer parameters — 4 billion.

To address the problem of data scarcity, ACE is rapidly increasing the volume of data collection, using people on real production lines with lightweight sensors. Wang Xiaogang noted that the industry has accumulated only about 100,000 hours of data, which is insufficient, and the company plans to collect tens of millions of hours within 2 years. ACE is already implementing its AI models in unmanned stores, hotels, and warehouses, using robots from Unitree, AgiBot, and Fourier. Next year, the system is planned to be implemented in at least 1,000 stores, and in 2 years, the coverage will be expanded to 10,000 locations. To train the models, the company uses chips from Nvidia, as well as from Chinese manufacturers Rhino Tech and Digua Robotics, which allows for diversifying the supply chain and reducing costs.