The "Water-Extraction" Process in the Embodied Intelligence Industry: Who's Paying the Price When Robots Start "Faking Sales"?
Hello everyone, I'm your financial observer. Today, we're going to discuss a topic that's been making a big splash in both the tech and financial circles—and it's even causing some heated debates: the suspicion of revenue fraud in the embodied intelligence industry (that is, humanoid robot industries).
In simple terms, some people are accusing robot companies of manipulating transactions with local governments and investors to make their financial reports look better by counting unsold robots as revenue. It's like a restaurant owner who lets relatives and friends come in for meals every day to create the illusion of a busy business, but the money is actually coming out of their own pocket.
The core conflict of this issue is that Shao Tianlan, the founder of Meikamand, has publicly named companies like Galaxy General for using "data collection centers" and related-party transactions to create fake revenue. Meanwhile, regulatory authorities (such as the China Securities Regulatory Commission, CSRC) have tightened the IPO (Initial Public Offering) requirements, demanding that companies prove they have genuine, recurring revenue.
Below, I'll break down this complex news into five easy-to-understand points to help you grasp the situation:
---
1. What is a "data collection center," and why has it become a hotbed for fraud?
First, let's clarify what the "data collection center" mentioned in the news is. You can think of it as a kind of "robot training school":
- Traditional approach: Robots need to watch a lot of videos and listen to a lot of sounds to learn.
- Embodied intelligence approach: Robots need to perform tasks (like opening packaging or tightening screws), so just watching videos isn't enough; they need to actually practice. Companies place hundreds or thousands of robots in factories, have them perform these tasks, and record the data to train AI models.
Why is this a problem?
Previously, robots were sold to factories, hospitals, or households—these were real market demands with actual cash payments. But now, many robots are sold to data collection centers, which are often established by local governments or in joint ventures with robot companies.
This creates an awkward situation:
- On the surface: A robot company sells 10,000 robots to a data collection center and reports hundreds of millions in revenue, making its financials look great.
- In reality: These robots don't go to factories; they just sit in the centers. The money for buying these robots often comes from local subsidies, state-owned platforms, or even the company's own shareholders.
It's like running a driving school where you let relatives and friends pay to enroll, but you're actually paying out of your own pocket or relying on government subsidies. Even if there's revenue on the books, it doesn't mean the school is truly popular or profitable.
---
2. How do related-party transactions inflate revenue?
The news mentions a key concept: related-party transactions. In business, if Company A sells something to Company B, and A and B are related (for example, B is a major shareholder of A or they're controlled by the same owner), it's considered a related-party transaction.
In the embodied intelligence industry, these transactions are common and mainly happen in two ways:
1. State-owned/government platforms paying: For instance, Zhiyuan Robotics won a project in Zhuhai, and the buyer was Zhuhai Zhihuiyuanqi Technology Co., Ltd., a joint venture between Zhiyuan and local state-owned assets.
- Implication: Although the money goes through official accounts, there might be government support or a form of subsidy behind it. If the robots are just displayed or used for data collection without generating real commercial value, the revenue is fake.
2. Investors as major customers: For example, Galaxy General's largest customer is CATL, which is also a major shareholder (lead investor in its Series A round).
- Implication: Investors may place orders to support the company's IPO or maintain its valuation. These orders are real, but they reflect the investors' desire to help the company, not market demand.
What do experts say?
Zhang Xinyuan, an expert from the Shanghai Technology Exchange Think Tank, suggests two indicators to judge the authenticity of such revenue:
- Proportion of revenue from related-party transactions: If 80% of revenue comes from related parties, the risk is high.
- Real repurchase rate after excluding related parties: Are there non-related buyers? Will they buy again next year? If not, it indicates the product may not solve real user problems.
---
3. Regulators stepping in: Higher IPO requirements, no more room for "stories"
This controversy has erupted because of the IPO process. Previously, the capital market liked stories like "We have humanoid robots that can replace human labor, with a valuation of tens of billions!" As long as the story sounded good, money would flow in. But now, the attitude has changed.
According to The Information, the CSRC has issued non-binding guidance to investment banks and firms, stating:
- If you want to go public, you need to prove you can generate sustainable revenue or at least show a narrowing of losses.
- You need to demonstrate real technological innovation, not just rely on PPTs and press releases.
- Your revenue must be recurring and not based on related-party transactions.
What does this mean?
The old model of burning money to expand or relying on government subsidies and shareholder support will face strict scrutiny during the IPO process. Regulators don't want companies that rely on fake sales to enter the stock market, as a drop in stock prices could harm retail investors.
The National Development and Reform Commission has also stated, "We need to prevent blind enthusiasm and encourage targeted, practical applications of robots." This signals a shift in policy, emphasizing that the industry must focus on practical solutions.
---
4. The current state of the industry: Half water, half fire
Despite the controversy, we can't dismiss the entire industry. Embodied intelligence is a trillion-dollar market, but its current state can be described as "half water, half fire":
Risks:
- Inflated revenue: Many companies' 2026 revenue includes 30%-50% or more from data collection centers and related parties. For example, a company valued at over 20 billion expects to sell 1,000-2,000 robots, with 500 from data collection centers generating 350 million in revenue.
- Lack of market relevance: Many robots can only mimic human actions (dance, shake hands), but they don't improve efficiency, increase costs, or are unstable in real factories.
- Low data transaction prices: A transaction for 300 pieces of data in Zigong costs only 30,000 yuan, indicating that data doesn't create high-value commercial cycles.
Opportunities:
- Infrastructure improvement: There are over 90 data collection centers across the country, accelerating data accumulation, which is essential for AI development.
- Technological progress: Although commercialization is challenging, technology is advancing. 2026 is a critical year for transitioning from concept validation to practical applications.
- Leading companies: Companies like UbiSelect, Zhiyuan, and Galaxy General, despite related-party transactions, are driving industry standards and technology implementation.
Key point: The industry is going through a process of "water extraction"—companies with real technology, practical applications, and repeatable sales will survive; those that rely on fake sales will be eliminated or delayed in getting listed.
---
5. Insights for investors:
If you're interested in tech investments or the robot industry, this controversy offers several important insights:
1. Don't just look at sales volume: Consider the repurchase rate. If a company sells 10,000 robots this year, that's impressive, but will customers buy next year? If the robots are all for data collection centers, will those centers still exist next year? If the customers are all related parties, is the demand sustainable?
2. Be wary of valuation bubbles: Many embodied intelligence companies' valuations (e.g., 20-30 billion) are based on future expectations, not current profitability. Tighter regulation could significantly reduce these valuations.
3. Focus on real applications: Don't just focus on robots dancing on TV shows or shaking hands at press conferences. Look at whether they can perform practical tasks in factories, warehouses, or hospitals. Only companies that can prove their models work in real-world scenarios are truly valuable.
4. Understand policy changes: The government no longer encourages the mass construction of robot parks but promotes targeted, practical applications. This means subsidies will likely go to companies that solve real industry problems.
In summary:
The embodied intelligence industry is moving from a turbulent "adolescence" to a more mature phase. Companies that rely on fake sales and exaggerated stories will face challenges. Those with real technology, practical applications, and profitability will have the opportunity to clear the clutter and establish themselves.
For us as consumers, it's important to see beyond the hype: future robots are meant to help humans, not just to perform for investors. Only those companies that can deliver real value will be the winners.