Chinese humanoid robots' order frenzy '

Just two days later, Ubiquitous announced another good news, reaching a 250 million yuan purchase contract for embodied intelligent humanoid robots with a leading domestic enterprise, delivering WalkerS2 as the main product, breaking the record for a single order from a global humanoid robot enterprise.

Looking back to the first half of 2025, there have been over 83 publicly disclosed projects for humanoid robots in China, with a total contract amount of nearly 330 million yuan. Only three companies, Ubiquitous, Yushu Technology, and Zhiyuan Robotics, have taken 60% of the market share.

However, behind the seemingly prosperous trend of "explosive orders" lies a deep-seated industry concern.

From the early years when new energy vehicle companies collectively skipped a billion yuan order due to immature technology and unsuitable scenarios, to the repeated delays in Tesla Optimus's plan to mass produce 5000 units by 2025, and to the technical barriers faced by the landing of the Robot World model, all indicate that the industry has not yet entered a mature stage.

For industries that are still in the stage of "not converging on the route", such "star events" are essential. They complete popular science popularization for the industry, pull talent, funding, and policy attention to the same track at once, and provide a story entry for subsequent "patient capital" to enter, making long money willing to stay.

Under the carnival, it is even more important to calm down and settle down. Eyeballs and orders are indispensable "fuel" for early industries, and what really needs to be wary of is mistaking the spotlight for the finish line and treating the story directly as a financial report.

In the embodied intelligence industry, there is still significant controversy over the technological roadmap.

Taking the field of embodied synthetic data as an example, there is currently a heated debate over the two main technical routes of "video synthesis+3D reconstruction" and "end-to-end 3D generation". The former involves creating videos or images, reconstructing them into 3D data, and ultimately transforming them into structured semantic models. This approach is adopted by projects such as Li Feifei's "World Model" project, as well as SpatialLM and SpatialVerse from Qunhe Technology. And "end-to-end 3D generation" directly synthesizes structured spatial data using graph neural networks, diffusion models, etc. Models such as ATISS and LEGO Net are representative of this route.

The current technological roadmap in the field of robotics is still in a divergent stage, and whether it is the overall structure, driving mode, control algorithm, or how to couple large models, it is far from being determined. Before the signal of convergence of the technological roadmap is clear, excessive resources should not be invested in a single technological roadmap to "feed" it.

Under various contradictory factors, the more fanatical the capital is, the more it must restrain the impulse of "one shuttle", and this consensus in the industry has become increasingly clear.

Because of this, mainstream institutions have quietly drawn two red lines internally: first, the number of members in a single team cannot exceed 100, and second, if the amount of a single round of financing does not match the current stage of the team, they will not invest. The purpose of doing this is actually very simple, which is to break down the limited trial and error resources into more "small samples", so that different technical routes can be promoted in parallel, rather than stuffing all the firewood into one stove and burning the "water" at the bottom of the pot early.