There has been a heated debate in the intelligent robot circle recently.
Shao Tianlan, the founder of Mecamand, which just went public shortly after, posted in Moments questioning the so-called "game-building" embodied intelligence entrepreneurship: "Nowadays, many 'game-building' embodied intelligence companies, including some well-known and highly valued companies in Beijing and Shanghai that participated in the Spring Festival Gala, make extensive use of so-called 'data mining centers' and related transactions with local governments, investors, and suppliers to create false and unsustainable income. This practice is illegal, unethical, and unsmart."
This statement ignited public opinion in the industry and put the data mining center at the forefront.
Many people's first reaction will be to ask, is the data mining center easily used to beautify financial reports? In fact, the data mining center should not be the "scapegoat" for this matter. We cannot simply put a negative label on the data mining center and characterize the data mining center as a financial operation tool.
One of the core bottlenecks of current embodied intelligence is the extreme lack of high-quality real-world data. The results of Stanford's "2026 AI Index Report" and multiple independent evaluations show that the success rate of embodied intelligent robots in simulation environments is as high as 89.4%, but in real home scenarios it plummets to 12.4%.
If a robot wants to have the ability to adapt to changing circumstances, it needs to be "fed" bit by bit with data obtained through training in real scenarios.
The data mining center is a data training ground for embodied intelligence, supporting embodied intelligence data production, model training, system application capability verification and iterative optimization in a real physical environment. Data collection personnel or robots will collect real physical interaction task data in different real scenes in the data mining center. These data will become "teaching materials", which will be bought back by robot companies and used for model training to promote the faster implementation of robot products in daily life.
Therefore, the necessity of the existence of a data mining center is self-evident.
According to incomplete statistics, as of the end of June 2026, more than 70 training grounds have been built and opened across the country, and 46 training grounds are under construction or planning. The logic of training ground construction is shifting from being driven by talent and capital to a resource allocation model centered on real scenarios, data supply and operational efficiency.
However, building the embodied intelligence training field is only the first step. What is more important is to use it and turn it into a real data resource base and real-life training platform to support the large-scale development of the embodied intelligence industry.
Therefore, the current key is not to deny the value of the data mining center. Real training, data collection, and scenario iteration should be supported.
Of course, recognizing the industrial value of data mining centers does not mean that the industry does not need to be wary of irregular transaction models. There is indeed a type of cooperation structure in the market: companies sell robots to data acquisition centers and promise to buy back the data.
This closed-loop trading model can easily trigger market discussions on the substance and commercial rationality of the transaction. However, related transactions only indicate the existence of a relationship between the two parties. Low income quality does not mean income fraud, and high business model risk does not mean illegality. Whether there are any problems with related-party transactions depends on whether there is real commercial delivery, whether the price is fair, whether there is circulation of funds, and whether the related-party relationships are fully disclosed.
After all, real-life scenario training, data collection, and product iteration are needed for industrial development; transaction arrangements that lack real business support require careful screening by all parties in the market. For contracts that lack real business support and have doubtful transaction purposes, regulatory agencies and intermediaries need to maintain prudent verification, and investors should also pay attention to business implementation capabilities, not just looking at paper revenue.
Data mining centers are "data granaries" and "real-life training grounds" for the development of embodied intelligence. The industry must support real collection, real training, and real delivery, and must also maintain penetrating supervision on related transactions, capital circulation, and income quality, so that data mining centers will not become "scapegoats."
The growth cycle of the hard technology industry is long, but the embodied intelligence track is currently in a sensitive period in the capital market, and the market is re-evaluating the valuation logic of the humanoid robot track. The capital market's patience with cutting-edge hard technologies is hard-won, and industry confidence requires multi-party care. For the embodied intelligence track, which is still in the early stages of industrialization, the more important thing is not to question each other, but to focus on the implementation of technology.
The embodied intelligence industry involves many cutting-edge technologies such as artificial intelligence, advanced manufacturing, and new materials, and is an important part of the future industry. The Shanghai Stock Exchange lists robots and embodied intelligence as key support directions for the Science and Technology Innovation Board, while insisting on strictly controlling listing access and maintaining the bottom line of business authenticity; the National Development and Reform Commission proposes to focus on practical application and implementation results, using the embodied intelligence training ground and application pilot base as the starting point to allow robots to iterate technologies in real scenarios and form an application closed loop around real needs.
At present, domestic companies have accumulated a lot of achievements in humanoid robot hardware and embodied models. Internal friction will bring obvious negative spillover effects to the industry. All parties in the industry should work together to improve scenario verification and data systems, jointly rebuild capital market confidence, and promote stable and long-term industrial development.
