You will have seen the videos of humanoid robots dancing, walking and doing backflips on social media. Behind most of them stands a single company: Unitree. According to Wired, the company's G1 and R1 models shipped more than 5,500 units in 2025, and it is preparing for a public listing in China.

Why they are so visible

Unitree's success is as much commercial as technical. Humanoid robots were for a long time million-dollar machines belonging to research laboratories. The G1 broke that pattern: a relatively affordable robot that walks out of the box and can be programmed.

That price point opened a new customer base — universities, small research groups, entertainment companies and content creators. That is where the viral videos come from: most people who buy the robot are not doing a job with it, they are showing it.

The real question

That is exactly the question Wired asks: the G1 has managed to charm a crowd, but can it ever hold down a real job?

In humanoid robotics that is the hardest part. A robot walking is a solved problem; a robot completing a useful task in an unpredictable environment is not. There is a long distance between walking straight on a factory floor and carrying boxes through a cluttered warehouse.

Examples of humanoid robots used in real jobs remain limited so far, and most are at pilot stage. A sales figure of 5,500 does not mean 5,500 robots are working.

The IPO context

The company preparing to list makes that question more pressing. A robotics company's valuation rests less on the units it sells than on the expectation of what those robots will be good for.

From an investor's view there are two scenarios:

  • A research platform — the robot remains a tool, with a customer base limited to labs and schools. The market is real but small.
  • Labour — the robot does work in a warehouse, a factory or a service setting. The market is very large but not yet technically proven.

Pricing follows the second scenario while today's usage sits closer to the first.

The AI side

The bottleneck here is not mechanical but software. The robot's legs work; what is missing is the layer that decides what to do. That makes it the physical counterpart of the same problem seen in language models: capability without reliability.

Most recent work in robotics targets exactly that gap — models combining vision, language and motion. When that layer settles, the figure of 5,500 will mean something very different.

Why China is ahead

Chinese manufacturers lead in humanoid robotics because of production rather than research. Most of a robot's cost sits in motors, gears, sensors and batteries — all parts of the electronics manufacturing chain.

Where that chain is concentrated, building a robot cheaply becomes possible. The same advantage played out earlier in drones and electric vehicles; it is repeating in humanoids.

Most Western rivals are developing more capable but far more expensive machines. Which approach wins depends on whether the robot can genuinely do a job: if it can, capability decides; if it cannot, price does.