Zuckerberg brand robot, made in China
Meta CEO Mark Zuckerberg. Image processed by AI
Written by｜Zhou Xiaoyan
Editor｜Xu Qingyang
Meta is entering the humanoid robot arena, with the first prototype developed by a Chinese company.
According to foreign media reports, Meta has collaborated with the Chinese robotics company Fourier, completing a humanoid robot prototype named Max in just three months, and has now initiated the development of the second generation product. This marks the most concrete hardware progress since Meta established its robotics team.
From AI models to humanoid robots, Zuckerberg is extending Meta's technological layout into the physical world. Notably, in this collaboration, Meta chose not to build a hardware system from scratch but instead leveraged the robotics research and manufacturing capabilities of the Chinese company to quickly complete the prototype development.
What tasks will this robot undertake? Why did Meta choose Fourier? Why can't American tech giants avoid Chinese manufacturing in their robotics layout?
### 01 Prototype Built in Three Months
According to foreign media reports, Max stands about 1.5 meters tall, weighs approximately 48 kilograms, and is equipped with an NVIDIA Jetson Thor chip. The robot has two built-in speakers in its chest, a face with a display that can simulate human expressions, and hands designed with five fingers, capable of picking up small bottles and toys.
Meta's requirements for this robot go beyond basic mobility; they also involve human-robot interaction. According to insiders, Max needs to be able to recognize sounds from five meters away, gestures and waving from three meters away, maintain eye contact, and express states such as listening, thinking, speaking, and uncertainty through different status signals.
This robot is customized based on Fourier's existing technological foundation. Meta utilized Fourier's previous hardware and research experience from the GR series of robots, and both parties subsequently adjusted the robot's height, body proportions, head structure, display, and camera, as well as modified the hands and shoes.
Zuckerberg with Fourier's GR-3 concept robot
Li-Chen Miller, Vice President of Meta Robotics Products, worked at Fourier's Shanghai factory for nearly two months, collaborating with the local team to advance research and development. At the end of August, Max was showcased to some employees at an internal Meta robotics summit.
In September, both parties initiated the development of Max 2. The new version plans to redesign the torso, improve hand operation capabilities, add a projector for displaying visual content, and equip additional sensors to help the robot navigate stairs.
A Meta spokesperson stated that the collaboration with Fourier is primarily for internal research and experimentation, and the company has no plans to sell these robots.
This does not mean that Meta does not have larger product intentions. Insiders indicate that the company is hiring supplier development engineers in Shanghai and product design engineers and technical project managers in Shenzhen, with work involving product design, manufacturing, prototype development, and preparation for mass production.
From having R&D personnel stationed in China to recruiting for supply chain positions, Meta is extending its robotics project from a single prototype to a more complete research and manufacturing system. However, there is currently no evidence that Meta has finalized mass production plans or product release timelines.
### 02 Giving Muse a "Body"
The core goal of Meta's development of Max is related to its AI agent Muse.
Meta's Chief Technology Officer Andrew Bosworth has referred to the humanoid robot internally as the physical embodiment of Muse. This means that Meta hopes AI can not only answer questions through conversation but also perform tasks in real-world environments using the robot.
In May of this year, Meta acquired the robotics AI startup Assured Robot Intelligence. This company focuses on developing robotic models that help machines understand human behavior and adapt to different environments. The relevant technology and team have also become part of Meta's robotics research and development.
From the current layout, Meta is filling in the software capabilities needed for robotic AI while testing how these capabilities integrate with hardware through Max.
For example, when a user asks the robot to clean the table, the AI must first recognize the items on the table, determine which ones need to be moved, and then plan the actions for grabbing and placing. During execution, the robot also needs to adjust its grip based on the position, weight, and shape of the items. If an item slips or the planned route is blocked, it must reassess its next action.
Such tasks involve visual recognition, spatial understanding, action planning, and motion control. Just because AI can understand a command does not mean the robot has the capability to execute it.
Meta needs a physical platform like Max to test how models perform in real-world environments. The perceptual errors, action failures, and environmental changes encountered by the robot while executing tasks can all serve as the basis for subsequent model training and hardware improvements.
Meta's robotics business has also undergone management adjustments. Marc Whitten, who previously served as CEO of General Motors' autonomous driving division Cruise, left at the end of August, just about 18 months after joining Meta. Subsequently, Meta appointed Ning Li as the interim head of Meta Robotics Studio.
Bosworth took over Whitten's previous responsibilities and continues to be involved in the technical and operational work of the department.
For Meta, whether Muse can evolve from a software agent into a system capable of executing real-world tasks depends on whether the model and robot hardware can work in harmony. Currently, Max is undertaking this research and development task.
### 03 The U.S. Does AI, China Does Hardware
Meta's choice of Fourier indicates the importance of Chinese robotics manufacturing capabilities in the global industrial chain.
Fourier initially focused on exoskeletons and rehabilitation robots before entering the humanoid robot field. Its GR series products have undergone multiple iterations, accumulating research and development experience in bipedal movement, dexterous hands, and tactile perception.
Promotional image of Fourier's GR-3 concept robot
For Meta, collaborating with a company that has existing products and manufacturing experience can reduce the exploration time from mechanical structures and components to complete assembly, allowing more resources to be invested in AI models and interaction systems.
Chinese companies also have advantages in industrial support and practical application scenarios. Humanoid robots involve multiple components such as motors, reducers, sensors, structural parts, and control systems. China has a relatively complete manufacturing supply chain and numerous factories and industrial clients that can provide testing environments.
According to foreign media reports, in the first half of 2026, Chinese embodied intelligence companies secured 93.5 billion yuan in financing across 322 transactions, approximately 13 billion dollars, five times that of the same period last year. The investment scope has extended from complete machine manufacturing to robotic AI models, training data, and software systems.
Chinese companies are attempting to solve the problem of how robots learn unfamiliar tasks.
Spirit AI has hired about 1,000 people to record daily life and industrial activities to accumulate data for robot training. ACE Robotics collects motion and grip information from store and hotel staff through wearable sensors.
The robotics division of Ant Group, Robbyant, claims that its model has used 160,000 hours of training material, including 50,000 hours of human-guided robot movement data, and has trained software on 20 robot configurations from 17 brands.
Another company, DexForce, uses computer-generated environments to train robots to reduce the cost of collecting real-world scene data.
These companies have different technical routes, but share a common goal: to enable robots to no longer rely entirely on action programs pre-written by engineers, but to adjust their behavior based on changes in the environment and tasks.
American companies are also advancing in this direction. Startups like Physical Intelligence, Skild AI, and Generalist are developing general robotic models, hoping that robots can learn multiple skills and reduce the need for repetitive training for single tasks.
The U.S. has advantages in certain advanced AI research and computing resources, while Chinese companies are closer to the manufacturing process and industrial application scenarios. The competition between companies in both countries in the robotics field has extended from hardware manufacturing to models, data, and computing platforms.
However, the collaboration between Meta and Fourier also faces policy restrictions.
In July, the U.S. Federal Communications Commission (FCC) restricted the import of new humanoid robots from China due to potential supply chain and cybersecurity risks. The regulations at that time allowed companies to import up to 4,000 devices for testing and product development, but prohibited sales. In August, the FCC proposed to reduce this exemption number to 40 devices unless companies obtain written approval.
If this proposal is implemented, American companies' ability to import robot prototypes from China for testing, as well as to continuously obtain relevant components, may be affected.
This creates a dual reality for Meta's collaboration: on one hand, Chinese companies can provide mature hardware research and manufacturing capabilities; on the other hand, U.S. restrictions on the Chinese robotics supply chain may increase subsequent collaboration costs.
However, for Meta, Max has not yet provided sufficient public data to demonstrate its actual capabilities. The robot's task completion rate, battery life, load capacity, and the amount of human assistance required during execution have not been disclosed. Adding hand and stair sensors does not equate to solving the complex object manipulation and stair navigation issues.
For home robots to truly enter the consumer market, reliability, safety, and cost issues need to be resolved. Consumers need devices that can consistently perform specific household chores, adapt to different home environments, and stop promptly in case of emergencies, rather than just machines that can understand commands and perform actions.
As AI begins to enter the physical world, the competitiveness of products will depend not only on how smart the model is but also on whether it can translate understanding into action through a reliable body. For the robotics industry, this is more challenging than simply manufacturing a humanoid robot that can walk.