Meta and Fourier Built a 48 kg Humanoid Named Max in Three Months, and Max 2 Is Already Underway

Meta Platforms has spent years assembling the pieces of a robotics program without ever showing the public a robot. It bought a San Diego startup that builds AI software for machines and folded that team into Meta Superintelligence Labs. It described humanoid robots as a possible body for Muse, the consumer AI agent the company has been building. What it had not done, at least publicly, was put a working machine on the floor.

That changed this week, not with a launch but with a prototype. The Information reported on October 31, 2026 that Meta built a humanoid robot with Shanghai-based Fourier Intelligence in about three months, and that a second version is already in development. The machine is known inside the company as Max. It stands roughly 1.5 metres tall, weighs about 48 kg, and runs on Nvidia's Jetson Thor chip.

The details are unusually specific for a project Meta has never announced. They include the size of the robot, the identity of its Chinese hardware partner, the name of the executive who relocated to work on it, and the features planned for the follow-up. They also include a limitation: Meta says the work is for internal research and experimentation, and that it has no plans to sell the units.

For a company of Meta's scale, a 48 kg prototype in a research lab is a small thing. The reporting around it is not. It places a Silicon Valley platform company inside a Chinese robotics supply chain at a moment when humanoid hardware is becoming the most contested layer of physical AI, and it shows how quickly a software-first company can buy its way into a working machine.

Key Facts

The Information's Jyoti Mann reported the project on Saturday, October 31, citing people familiar with it. Reuters reported on October 31, 2026 that Meta has built a prototype humanoid with Chinese startup Fourier and is working on a second version, attributing the account to The Information, and was explicit that it could not immediately verify the report. The wire dispatch carried no specifications, no development timeline, no estimated cost and no comment from either company.

Investing.com reported on October 31, 2026 that the internally named Max weighs about 106 pounds, or roughly 48 kg, stands around 1.5 metres, and uses Nvidia's Jetson Thor. The robot has a display that mimics facial expressions, built-in speakers and five-fingered hands designed to pick up small objects. The Next Web reported on November 1, 2026 that the two companies built the prototype in about three months and that a second version is already underway, while noting it had not independently verified the report.

Humanoids Daily reported on November 1, 2026 that Meta's requirements for the machine emphasised interaction with people. Those requirements reportedly included recognising a voice from five metres away, interpreting a wave or gesture from three metres, maintaining eye contact, and signalling whether the robot is listening, thinking, speaking or uncertain. The same report described hands able to pick up small bottles and toys, chest-mounted speakers and a face that can display expressions.

People matter as much as parts in the account. Meta's robotics team travelled to China to work with Fourier, and Li-Chen Miller, Meta's vice president of product for robotics, spent almost two months at Fourier's factory in Shanghai. The collaboration culminated in an internal demonstration at a Meta robotics summit in late August 2026. Work on Max 2 began in September 2026, with a redesigned torso, improved hands, a projector and additional sensors intended to help the machine handle stairs.

Meta's public position is narrow. A Meta spokesperson said the effort with Fourier is focused on internal research and experimentation, and that the company has no plans to sell the units. The Information reported that Meta has not set a launch date for a humanoid product, according to one source. Fourier, based in Shanghai, unveiled its GR-3 care-focused humanoid in August 2025; that machine stands about 1.65 metres, weighs about 71 kg and offers up to 55 degrees of freedom, which makes Max a smaller and lighter design.

Analysis

The obvious reading is that Meta is late to the hardware side of this race. Its own public output has been software and research papers rather than machines, while hardware specialists including Fourier have been iterating on humanoids for years. Building a prototype in roughly three months, with an outside partner doing the manufacturing, is less a feat of engineering than a feat of sourcing. What this really means is that Meta has decided the fastest route into embodied AI is to rent a robot body and concentrate its own effort on the model inside it.

That division of labour is the most interesting part of the story. Fourier has spent years building rehabilitation robots and its GR humanoid series, while Meta has spent years building models, agents and interaction design. AI Market Watch reported on November 1, 2026 that the arrangement illustrates one possible split in physical AI, where a software company uses a specialist's hardware to explore embodied systems. If that split holds, the competitive advantage in humanoids may sit less with whoever welds the chassis and more with whoever owns the interaction layer, the data and the agent that gives the machine a reason to move.

The interaction requirements reported by Humanoids Daily point the same way. Recognising a voice from five metres, reading a wave from three metres, holding eye contact and signalling internal state are not locomotion problems. They are product problems, and they are the kind of problems a consumer AI company tends to treat as the real work rather than an afterthought. Meta's CTO Andrew Bosworth has internally described humanoids as a physical embodiment of Muse, the consumer AI agent, according to the report. A robot that can indicate it is listening or uncertain is a robot designed by people who think about interface states, not torque curves.

The caveats deserve equal weight. Reuters reported on October 31, 2026 that it could not immediately verify the account, and The Next Web said the same of its own reporting. The core claims rest on people familiar with the project speaking to a single publication, with later outlets relaying rather than independently confirming. Nobody has published a photograph, a demonstration video or a benchmark. AI Market Watch noted that an internal prototype does not establish commercial adoption, and that the next meaningful evidence would be demonstrated performance on manipulation and navigation rather than the speed of prototype assembly. A three-month build is a schedule, not a capability.

Why It Matters

The strategic context is a race that has moved from software to bodies. Meta has said little in public about selling a robot, but it has been buying the ingredients. It acquired Assured Robot Intelligence, a San Diego startup developing AI models that help robots understand and adapt to human behaviour, in May 2026, and that team joined Meta Superintelligence Labs. The Next Web reported on November 1, 2026 that Nvidia, whose chip runs Max, agreed the same week to invest 3 billion dollars in Meta and expand their work on Muse, a figure that underscores how much capital is circling the intersection of agents and machines.

There is also a geographic story. A flagship American platform company chose a Shanghai hardware partner for its first humanoid, sent its robotics team to China, and left a vice president there for almost two months. That is a practical admission about where affordable humanoid engineering currently lives. The arrangement may be research-only, but supply chains have a way of becoming dependencies.

For consumers, the honest answer is that nothing changes yet. Meta has no launch date, no price and no plan to sell the units, and Max 2 is a specification list rather than a shipped product. The significance is directional. A large platform company has started treating a walking, talking machine as a plausible endpoint for its AI assistant, and it has found a way to get there without building a factory.

Next Up

The near-term test is Max 2. A redesigned torso, better hands, a projector and sensors for stairs are modest upgrades, and they suggest a team iterating on interaction and mobility rather than chasing a product deadline. Watch for any demonstration, filmed or staged, that shows the machine manipulating objects or navigating a real room. That, not another report about assembly speed, would be the first hard evidence.

The larger question is whether Meta ever says more. For now the record consists of a single original report, a wire service that could not verify it, and a company statement that narrows the project to internal research. Meta has not set a launch date, one source told The Information. Until it does, Max remains a prototype with a name, a weight and a chip, and a clear signal about where Meta thinks its AI agent needs to go next.