The "Hand-to-Hand" Game in the Robotics Industry: An In-depth Analysis of Data, Orders, and Real Demand
As a financial journalist who has long observed the technology industry and capital markets, my biggest takeaway from this report on the embodied intelligence (humanoid robots) sector is that under the surface of all the excitement, there is a sense of unease and concern.
In September, the embodied intelligence sector seemed to be in the midst of a frenzy of company listings and billions in orders. However, beneath this facade lies a reality that is embarrassing for both investors and industry players: money is circulating within the same circles, while the robots themselves are not actually being put into use in factories, and data is simply being traded back and forth.
The core question this article raises is: To whom are these expensive robots actually being sold? Who is truly paying for them?
To make this clearer, I have broken down the report into five key aspects and explained the logic, risks, and underlying issues in plain language.
---
The Embarrassing Reality: Only 1 Out of 5 Robots Is Actually Being Used for Work
First, we need to dispel a misconception: humanoid robots are not being mass-produced and purchased by consumers or factories in the same way that smartphones or cars are.
Here are some striking figures from the report:
- By 2025, approximately 18,000 humanoid robots are expected to be shipped globally, with China accounting for the majority (14,400 units).
- However, where have these robots gone? 78.4% of them have ended up in data collection centers, research institutions, and entertainment performances.
- A very small portion have made it into real production scenarios such as intelligent manufacturing and logistics.
In simple terms: Imagine a car company that produces 100 cars, but 90 of them are used for crash testing, with only 10 sold to the public. Crash testing is essential for safety, but if a company's revenue mainly comes from buying its own cars for testing, its business model is unsustainable.
This is exactly the situation in the current embodied intelligence industry. The so-called boom in shipments is largely an internal celebration within the industry. Robots are not being used in factories to perform actual tasks like screwing bolts or helping the elderly in homes; instead, they are gathered in training centers, practicing tasks like folding clothes and carrying cups.
Conclusion: The current market demand is more of a pseudo-demand or intermediate demand, rather than genuine end-user needs.
---
Who Are the Major Customers? Data Collection Centers Have Become the “Rustic Homes” for Robots
If factories are not buying robots, then who is? The answer is data collection centers. As of April this year, there were 64 such centers in operation nationwide, with over 90 more under construction, each equipped with more than 2,500 humanoid robots.
- Shanghai Zhiyuan has deployed around 200 robots.
- Cities like Zigong in Sichuan and Huizhou in Guangdong also have hundreds of robots.
What do these robots do there? They are not working; they are being trained. They carry cups in simulated kitchens, fold clothes in custom-made homes, and screw bolts on virtual production lines. Cameras record the precise movements of their joints, and this data is then packaged and sold as a product.
Why are data collection centers necessary? Robots lack the experience and knowledge needed to perform real tasks effectively. AI models like ChatGPT can learn from massive amounts of text online, but robots need real-world data to improve their capabilities. The industry consensus is that robots need tens of millions of hours of useful operational data to become commercially viable, yet the world currently only has a few hundred thousand hours of such data.
In simple terms: Data collection centers act as training grounds and “fitness centers” for robots. Local governments provide the venues, and robot companies provide the robots. Without these orders, many startups would struggle to continue their development.
Risk: This is similar to the early days of the internet, when infrastructure (such as cell towers) and electric vehicle charging stations) were built. If data collection centers become the only customers, even if the robots are capable, they will remain unused, creating a bubble in the industry.
---
The Financial Circulation: A “Hand-to-Hand” Financial Magic Show
This is the most critical and sensitive part of the report: How does the money flow within the industry?
Let’s look at the order patterns of leading companies:
- Youbixuan: Of the 11 large orders in 2025 worth tens of millions each, 7 went to data collection centers. In the last three months of 2025 alone, the company signed 6 orders worth approximately 829 million yuan, accounting for 60% of its annual total.
- Zhiyuan, Galaxy General, Leju Intelligence, and others have also won contracts for data collection centers, training facilities, and research platforms.
Here’s the sequence:
1. Step 1: Robot companies sell robots to data collection centers (often state-owned or government-owned entities), with money flowing from the government to the companies.
2. Step 2: The data collection centers use these robots to collect data.
3. Step 3: The robot companies buy back the data to train their algorithms, with money flowing back from the centers to the companies.
In simple terms: Suppose you open a fitness center and buy 10 treadmills from a neighboring company. The company earns money from the sales, and your revenue increases. Then you buy data usage services from the company, and your revenue increases again. The result is that both companies appear to be profitable, but no real value is created.
This is what some critics, like Shao Tianlan, refer to as a “bubble-building” entrepreneurial model: Local governments want to see industrial development and economic growth; companies need revenue and a higher valuation for listing or financing; investors want attractive financial reports; suppliers need orders. All these interests intersect to create a cycle where no one is paying for the actual use of the robots.
---
Regulatory Actions: IPOs Tightened, Pinning the Bubble
Just as the industry was self-promoting, external regulators stepped in. According to foreign media reports, the CSRC (China Securities Regulatory Commission) is considering stricter regulations for humanoid robot companies’ IPOs, requiring recurring revenue, reduced losses, and real innovation capabilities.
In simple terms: In the past, companies could go public as long as their stories sounded good and their presentations were impressive, even if they were losing money as long as their revenue figures looked positive (often through internal transactions). Now, regulators are asking: Is your revenue genuine? Are your customers real? Is your technology truly innovative or just riding on a trend?
This move targets companies that rely on one-time large orders, such as those selling to government data collection centers. If the door to IPOs closes or the requirements become more stringent, these companies will face significant financing challenges. A break in the funding chain could burst the bubble.
Galaxy General’s response highlights its focus on real research and development, core technologies, real use cases, and a genuine business model, trying to differentiate itself from companies that are merely creating artificial demand.
---
The Future Outlook: From a “Greenhouse” to the “Wilderness”
Finally, we need to consider the industry’s future:
The answer is yes, but the path must change.
1. Data is essential, but it must come from real-world scenarios. Data collected in simulation environments is useful, but it cannot replace the complexity of real-world situations.
2. The industry needs to move from an internal cycle to a positive cycle where factories see the practical benefits of robots and are willing to pay for them.
Key factors for success:
- Recurring revenue from actual customers, not just one-time orders.
- True innovation that provides a competitive edge in specific applications.
- A gradual transition where robots can replace humans in meaningful tasks and become more affordable.
Conclusion: The embodied intelligence industry has a huge potential, but it is currently in a chaotic phase. In the short term, companies that rely on artificial demand will be eliminated. In the medium term, those with genuine technologies and applications will survive. In the long term, robots will become more widely adopted when they can perform complex tasks at an acceptable cost.
For everyone:
- Investors: Be cautious of companies whose revenue comes mainly from government projects and data collection centers. Look at the proportion of their business coming from real end-users and the rate of product repurchases.
- Industry players: Focus on solving real problems in real-world scenarios, not just on increasing shipments.
- The public: Don’t be misled by headlines about robots entering factories. They are still mostly in training stages, far from being used in practical work.
In summary: The embodied intelligence industry is going through a period of purification. The days of money circulating within closed loops are coming to an end. The real test will be whether robots can prove their value in the real world.