Meta taps Fourier to build robot bodies, while Unitree seeks global partners for robot brains

Reading the two pieces of news from the past couple of days together is quite interesting—it also reflects the respective strengths and weaknesses of China and the US in the field of embodied intelligence.

The first: on the 2nd, US media reported that Meta has engaged Fourier to jointly develop a robot prototype, Max.

The second: on the 3rd, at the Wuzhen Internet Conference, Wang Xingxing said that the core focus for the next stage—the robot brain—is open to global cooperation.

First, let's talk about Meta engaging Fourier to jointly develop a robot prototype.

According to reports by The Information and other media, the robot Meta commissioned Fourier to develop is named Max; it is being advanced by Li-Chen Miller, Meta's VP of Product for its robotics business, working together at Fourier's factory in Shanghai, where Miller worked for nearly two months.

Meta's custom robot Max, developed with Fourier, is positioned as a consumer-grade robot capable of helping users with household chores and daily tasks. Meta CTO Andrew Bosworth internally described the humanoid robot as the physical embodiment of the company's flagship personal agent product "Muse."

The first-generation Max is about 1.5 meters tall and weighs about 48 kg.

It is equipped with Nvidia's Jetson Thor chip and a software platform that helps the robot move; two speakers are installed in its chest, the eyes and mouth on its face can simulate human expressions, and its five-fingered robotic hands can pick up small bottles and toys. It can recognize sounds from 5 meters away, understand gestures or waving motions from 3 meters away, maintain eye contact, and convey states such as listening, thinking, speaking, and uncertainty through different signals.

The first-generation Max was shown to some employees at a Meta internal robotics summit in late August.

Subsequently, in September, Meta and Fourier began development of the second-generation Max, namely Max 2. Max 2 will redesign the robot's torso and improve the robotic hands to enhance manipulation capabilities; it will also be equipped with a projector for displaying visual content. In addition, more sensors will be added to help the robot go up and down stairs.

Asked about developing robots with a Chinese company, a Meta spokesperson told The Information that it faces the same problem as other companies in the industry—difficulty obtaining robot components within the US—but that the focus of this cooperation with Fourier is internal research and experimentation, and there are no plans to sell these robots.

However, according to a report by Zhidongxi, Meta is also building an internal robot manufacturing team. Job postings show that Meta is recruiting supplier development engineers in Shanghai, responsible for working with product design and manufacturing teams to help the robotics business move from the prototype stage to large-scale production. At the same time, Meta is also recruiting product design engineers and technical project managers in Shenzhen, responsible for prototype development, project planning, and large-scale production for Meta Robotics Studio.

In addition to advancing robot R&D on the hardware side, Meta also acquired Assured Robotics Intelligence, a startup developing robot training models, in May on the robot control software side.

From this perspective, Meta, following OpenAI, is also determined to launch its own humanoid robot and, ignoring the FCC ban, has come to China to seek cooperation with Chinese companies.

The second piece of news is that on the 3rd, at the Wuzhen Internet Conference, Unitree's Wang Xingxing said that the core focus for the next stage—the robot brain—is open to global cooperation.

At the World Internet Conference Wuzhen Summit on the 3rd, Wang Xingxing said: Unitree has basically gotten its hardware capabilities—robot body, motion control, actuators, and so on—working, and the core focus for the next stage is the robot brain (that is, embodied intelligence large models).

To solve this global challenge of embodied intelligence, Unitree will open up to the world to seek cooperation, and is willing to open its robot hardware platform and motion control interfaces to carry out cooperation with large-model companies worldwide (the leading large-model companies are still American ones, such as OpenAI, Anthropic, Google, Meta, etc.), AI labs, universities, and others, jointly training and iterating robot large models to build the robot brain.

At this point, Unitree's robot hardware platform previously sold to developers has a major advantage: users who have already purchased it can directly raise requirements and work with Unitree to develop the robot brain.

This may be because, after testing GPT-6 Astra, Unitree found it could perform some actions without training, which made Unitree deeply aware of the capability boundaries of large models, and therefore it is seeking cooperation with global large-model companies.

Training a robot brain requires integrating multimodal data such as vision, touch, and proprioception, and performing planning and decision-making in the real physical world; it requires a large amount of physical interaction data, which is also the biggest bottleneck in the industry today. Therefore, companies that solve this data demand are sought after by capital. Jianzhi Robotics, which aims to produce embodied intelligence data, has announced the completion of $100 million in financing, completing six rounds totaling more than $300 million in just 14 months.

More importantly, Wang Xingxing judges that large models could crack the problem of embodied intelligence in 1-2 years, and believes that GPT-7, to be launched next year, may achieve the GPT moment of embodied intelligence.

This has given Unitree a sense of crisis, because large-model companies are indeed entering the field of embodied intelligence on a large scale.

Although it raised 6.1 billion yuan in its August 2026 IPO, more than 2 billion yuan of which will be invested in the robot brain, it is obvious that this money is far from enough compared with the financing of large-model companies—it can't even buy many cards.

With embodied intelligence companies lagging large-model companies by at least two orders of magnitude in both funding and computing power, Unitree Technology also sees that large-model companies such as OpenAI and Meta are all going to enter the field of embodied intelligence and build their own bodies. It would be better to take the initiative, using the hardware capabilities it holds—robot body capabilities, motion control capabilities, actuator capabilities, and robot cerebellum capabilities—to cooperate with large-model companies. This would both preserve its own room to survive and accelerate the industry's progress.

From Meta coming to China to develop robot bodies to Unitree seeking robot brain cooperation globally, this also reflects the biggest respective advantages of China and the US in the field of embodied intelligence.

US companies, relying on advanced computing chips, have advantages in computing power and funding, and lead in the large-model field; China, relying on its manufacturing capabilities, has a hardware advantage. If the respective advantageous parts of the two countries can cooperate, it would indeed accelerate the industry's progress.

This article is from the WeChat public account "Chezhi" (ID: invehc), author: Michael Yuan, published by 36Kr with authorization.

Although it raised 6.1 billion yuan in its August 2026 IPO, more than 2 billion yuan of which will be invested in the robot brain. But it is obvious that this money is far from enough compared with the financing of large-model companies—it can't even buy many cards.

With embodied intelligence companies lagging large-model companies by at least two orders of magnitude in both funding and computing power, Unitree Technology also sees that large-model companies such as OpenAI and Meta are all going to enter the field of embodied intelligence and build their own bodies. It would be better to take the initiative, using the hardware capabilities it holds—robot body capabilities, motion control capabilities, actuator capabilities, and robot cerebellum capabilities—to cooperate with large-model companies. This would both preserve its own room to survive and accelerate the industry's progress.

From Meta coming to China to develop robot bodies to Unitree seeking robot brain cooperation globally, this also reflects the biggest respective advantages of China and the US in the field of embodied intelligence.

US companies, relying on advanced computing chips, have advantages in computing power and funding, and lead in the large-model field; China, relying on its manufacturing capabilities, has a hardware advantage. If the respective advantageous parts of the two countries can cooperate, it would indeed accelerate the industry's progress.

This article is from the WeChat public account "Chezhi" (ID: invehc), author: Michael Yuan, published by 36Kr with authorization.