China’s Tech Giants Shift the AI Battlefield Into Robotics
Humans are building machines that look like humans and may eventually replace them in performing tasks currently done by people.
Investors are now treating embodied AI and autonomous agents as some of the most serious growth engines in artificial intelligence. UBS (Union Bank of Switzerland) sees capital flowing toward a new frontier, with Chinese tech firms racing to embed advanced AI models into robots and shifting the generative AI battlefield from chatbots toward physical autonomous systems.
Alibaba launched Qwen3.7-Max, a cutting-edge multimodal reasoning model, last week, distinguishing it with its tool-calling architecture. This digital brain orchestrates hardware, enabling robots to navigate, avoid obstacles, and plan tasks without a human operator. The company also released supporting robotic models, including a gripper agent and a vision-language system designed for real-world interaction.
Earlier this month, embodied AI startup Zeroth announced that its M1 humanoid, a mass-produced bipedal robot, had integrated Tencent's OpenClaw AI agent framework. A large language model can hear human speech, interpret intent, and instantly convert it into robotic movement, bridging cognitive intelligence and physical action. As Wu Bangyi, chief data officer at Tianyu Shuke, noted, language model development has largely focused on the digital realm.
Goldman Sachs warned that high-quality real-world data is the new gold and is in desperately short supply. AgiBot co-founder Yao Maoqing quantified the gap: while GPT-5 trained on roughly 10 billion hours of data, the entire robotics industry has access to only about 500,000 hours of quality embodied data. To address this shortage, X Square Robot partnered with home-services platform 58 Daojia to deploy cleaning robots into residential settings in Beijing and Shenzhen, using real homes as data farms. According to a report by the China Academy of Information and Communications Technology and Tsinghua University, nearly 30 training facilities and data centers for embodied AI have been built or approved across the country.
Posts
