Physical AI is becoming the new battleground for global tech giants. Recently, OpenAI, NVIDIA, and Tesla have made significant moves in the field of embodied intelligence, indicating that competition in the robotics industry has shifted from pure hardware manufacturing to the development of underlying infrastructure and industry standards.

For a long time, the humanoid robot sector was dominated by startups and traditional robotics companies. However, in the past six months, tech giants have begun to get deeply involved. OpenAI recently officially established a new team called "OpenAI Robotics" and has been actively recruiting core talents in areas such as simulation environments, data collection, and electrical engineering worldwide. It aims to define the standards for physical AI through the collaborative design of the brain and hardware.

image.png

Differing from OpenAI's full-stack evolution approach, NVIDIA has chosen to continue consolidating its dominant position in computing power and development platforms. By integrating the Isaac humanoid robot reference platform, Jetson Thor computing platform, Cosmos world foundational model, and Omniverse digital twin platform, NVIDIA is striving to build a "CUDA" ecosystem for robots. Its core goal is to have robot companies around the world complete training and testing on its platform, thereby establishing insurmountable development rules in the era of physical AI.

At the same time, Tesla is moving the Optimus humanoid robot from the laboratory to large-scale industrial mass production. Tesla has planned to shut down some car production lines and completely transform them into production bases for Optimus. Leveraging its own expertise in algorithms, self-developed chips, and automotive-grade supply chain, Tesla aims to reshape the cost and data feedback loop in the robotics industry through mass production.

In this technological transformation, the development of robotics in China and the United States has shown different paths. Domestic companies generally adopt a more practical approach, relying on their supply chain advantages and rich industrial scenarios to directly enter the front lines of automobile factories and logistics warehouses with high-cost-effective products, taking an "inverse path" from industrial terminals to scale manufacturing to accumulate data. In contrast, U.S. companies tend to establish rules in the virtual world, using powerful cloud computing capabilities to synthesize data for closed-door training.