The social media giant's robotics push leans heavily on AI software, but a reported China partnership raises the stakes
Meta is reportedly building a humanoid robot prototype with a Chinese startup. The company that made its name on photos of your cousin’s vacation now wants a body to go with its brain.
The claim surfaced in a post on X. It has not been independently confirmed. Still, it fits a pattern Meta has been building toward for more than a year.
What Meta is actually building
Meta’s push into humanoid robotics started in early 2025 inside Reality Labs. That is the division best known for VR headsets and smart glasses.
An internal memo from February 2025 laid out plans for a dedicated robotics group. Marc Whitten was tapped to lead it.
The focus is not what you might expect. Instead of mass-producing its own robots, Meta has prioritized software, sensors and AI.
The engine behind that plan is Llama, Meta’s family of AI models. The goal is consumer-facing robots that run on Meta’s intelligence, regardless of whose metal is wrapped around it.
There is also an internal project called “Metabot.” It draws on engineering talent Meta has recruited and targets models that improve a robot’s dexterity and general functionality.
The Chinese connection
The most eye-catching part of the story is the reported partner. Meta has held early-stage discussions with Unitree Robotics and other firms. Unitree is one of China’s more prominent humanoid robot makers.
Those talks, however, do not point to Meta-branded robots rolling off an assembly line anytime soon. Speculation that Meta and Unitree are jointly building a humanoid prototype remains unverified. The conversations so far appear to lean more toward software partnerships than co-designed hardware.
That distinction matters. Licensing AI to run on someone else’s robot is a very different business from co-developing the robot itself.
Buying a head start
Meta has also been shopping for expertise. In May 2026, it acquired Assured Robot Intelligence, known as ARI.
ARI is a US-based startup that builds robotic AI foundation models for humanoid control. A foundation model is a large, general-purpose AI system that can be adapted to many tasks.
The ARI team, including co-founders Lerrel Pinto and Xiaolong Wang, has joined Meta’s Superintelligence Labs.
Background: a crowded race
Meta is not entering an empty arena. Tesla has been pushing hard on humanoid robots. Chinese firms, including Unitree and UBTech, have been advancing quickly.
What this means
For Meta, the strategy is a hedge with a clear logic. Hardware is capital-intensive, slow to scale and full of supply chain headaches. By concentrating on AI models, sensors and platforms, Meta can stay in the humanoid race without betting the company on building robots itself.
The China angle adds another layer. Any deep hardware collaboration between a US tech heavyweight and a Chinese robotics maker could invite scrutiny over technology transfer and data. Whether the relationship stays at the software-partnership level or moves into joint prototypes will shape how much of that attention Meta draws.
For investors, the ARI deal and the Metabot project suggest Meta is treating robotics as a long-term AI bet rather than a near-term product line. Meta has not set out a timeline for consumer robots.
What to watch next: confirmation, or denial, of the reported prototype partnership. Signs of whether Llama-based models start appearing in third-party robots. And whether Meta’s robotics group, under Whitten, begins showing off working demos rather than memos.
