The competition in the physical artificial intelligence (AI) market is shifting from hardware to software development, focusing on creating the "brain" of robots. As merely enhancing the appearance and operation of robots proves insufficient for adapting to various work environments, experts argue that without AI capable of controlling both the movements and behaviors of robots, advancements will remain at the level of existing collaborative robots.
Industry insiders assert that the ultimate goal of physical AI is to develop a universal AI model, or "AI brain," that can be integrated into any type of robot to unify cognition, judgment, and behavior.
On October 20, market research firm IDC projected that global shipments of humanoid robots will exceed 510,000 units by 2030. According to Fortune Business Insights, the humanoid robot market is expected to grow from $6.24 billion this year to $165.13 billion by 2034.
As the market expands, the focus of competition is shifting from the design and pricing of robots to the AI that powers them. South Korea, the United States, and China are engaged in a tight technological race, each leveraging their strengths: South Korea's manufacturing capabilities, the U.S.'s advanced AI development, and China's extensive robot ecosystem.
South Korea is concentrating on securing specialized physical AI that can be directly applied in industrial settings. The government is promoting the "Physical AI Alliance," connecting robot manufacturers, AI companies, research institutions, and end-users to develop specialized models for industries such as manufacturing and logistics, and to validate them in real-world applications.
In collaboration with local manufacturers in Gyeongnam and Jeonbuk, the government is working on projects to apply AI robots in production processes to gather operational data. On the corporate front, Rainbow Robotics has introduced the "RB-Y1," a wheeled platform combined with robotic arms, designed for precision tasks in industrial environments.
The United States is pursuing a strategy that combines generative AI with robotics to enhance manufacturing competitiveness and national security. In July, the White House identified robotics, automation, and AI as key priorities in its research and development budget for the 2028 fiscal year, viewing robots as strategic assets that influence supply chain resilience and technological security.
In the private sector, companies are focusing on developing AI models to control robots. NVIDIA has unveiled "GRooT," a foundation model and development platform for humanoid robots. Google DeepMind has introduced "Gemini Robotics," a model that integrates visual, language, and behavioral capabilities, and is advancing it with "Gemini Robotics 2" and "Gemini Robotics On-Device," which operates directly on devices.
China is responding to the physical AI market by supplying low-cost hardware in large quantities to accumulate diverse field data. According to the National Development and Reform Commission (NDRC), the number of humanoid robot companies in China has surpassed 150. The Chinese government is promoting the standardization of related technologies while expanding applications in manufacturing, logistics, and services.
A notable example of affordable hardware is "Unitree," which sells its humanoid robot "G1" for $13,500, significantly lowering the entry barrier for humanoid robots. The G1 showcases dynamic movements such as walking, running, and performing flips, emphasizing its physical control capabilities. Unitree is expanding the physical AI ecosystem by offering robots at various price points and for different purposes.
Industry experts believe that the outcome of the physical AI competition hinges on which country secures the "universal AI brain" for robots first. The World Economic Forum (WEF) and Boston Consulting Group (BCG) analyzed in their report, "Physical AI: Opening a New Era of Industrial Operations," that "robotics foundation models are the cognitive core that enables universal robots, like humanoids, to flexibly perform multiple tasks in various environments without the need for reprogramming."
* This article has been translated by AI.
Copyright ⓒ Aju Press All rights reserved.