Summary in Plain Language
Yusu Technology, once the "first stock in the humanoid robotics sector" that gained attention thanks to the support of Steve Jobs and Lenovo's quick investment, saw its stock price plummet from a high of 1,100 yuan shortly after going public. In the first half of this year, the entire embodied intelligence industry saw a five-fold increase in funding, with a total of 93.5 billion yuan poured into the sector, leading to the emergence of 22 robotics companies valued at over 10 billion yuan. However, the vast majority of companies in this industry are losing money despite spending heavily. Many of the so-called "commercial orders" they publicize are merely for show and not actually used to replace human labor. On the contrary, it's small companies that don't manufacture robots or develop large models but hire people to perform actions and provide data for the robots that are now in high demand. These companies have a 30% gross margin, which is higher than most traditional industries, making them the first group in the entire chain to actually make money.
Detailed Analysis
1. 9.3 Billion Yuan in Hot Money Flowed to Data Collectors Instead of Robotics Companies
Many expected that companies producing the robots and the hardware components (such as joints and motors) would be the first to profit from the investment in embodied intelligence. However, the reality is quite the opposite: the industry is facing a severe shortage of data. There are only 500,000 hours of real-world interaction data that meet the industry's standards, while robots need tens of millions of hours of data to function effectively, leaving a gap of over 99%. Even though the valuation of robotics companies has soared to several billion yuan, they lack the necessary training data, which prevents their robots from becoming intelligent. As a result, a large amount of hot money has flowed from the primary market financing into data procurement, with companies paying for this data. This has created a virtually uncompetitive market where demand far exceeds supply. Small teams that previously worked on autonomous driving data annotation have quickly expanded to collect data for embodied intelligence and can set up bases in third- and fourth-tier cities within two months, with no shortage of orders. In essence, the industry is paying in advance for the services that will enable robots to perform tasks.
2. The Business Is Essentially About Leveraging the Wage Gap Between Urban and Rural Areas
The data collection business operates on a simple "contractor" model:
- Robotics companies offer around 100 yuan per hour for data related to robot movements.
- Middle-level contractors then subcontract this work to data collection companies, charging 70-90 yuan per hour.
- Finally, the data collection companies pay their workers 40-70 yuan per hour.
To maximize profits, these companies avoid setting up bases in major cities like Beijing, Shanghai, Guangzhou, and Shenzhen and instead move to cheaper cities like Suqian and Nanyang, where labor costs are much lower. Hiring new workers is easy—anyone who can read and write, regardless of education or gender, can start working after three days, earning a daily salary of 160 yuan, which is higher than the income of local supermarket clerks or tea shop employees. The only challenge is managing the workers, as they perform repetitive tasks that can be very monotonous. Many workers leave after a few days. Even JD.com's plan to establish a large-scale data collection base in Suqian has not progressed as expected due to difficulties in personnel management.
3. An Odd Phenomenon: The Smarter the Robots, the More Humans Are Needed
There is a widespread consensus in the industry that sufficient data is essential for robots to become intelligent. Overseas teams have used millions of hours of human-generated data to significantly improve the success rate of robots in manufacturing scenarios. Now, upstream companies require millions of hours of data and are willing to spend more on data acquisition rather than worry about waste. This has led to a situation where the more intelligent the robots, the more humans are needed to train them. The cost of usable data for robots has risen to 300-500 yuan per hour, and for high-demand scenarios, it can exceed 1,600 yuan per hour—more than what a typical worker earns in a week. The industry is still struggling to gather enough data for various use cases, often leading to multiple robotics companies sharing the same facilities to reduce costs.
4. A Circular Economy of Fake Orders to Inflate Valuations
The current profit model in the embodied intelligence sector does not involve selling robots to customers but rather relies on a hidden cycle: local authorities purchase products from local robotics companies to meet performance targets, then use these data to inflate the companies' valuations in the capital market, allowing them to raise more funds in subsequent rounds of financing. This is similar to the AI industry, where capital invests in large model companies, which in turn use the funds to buy NVIDIA graphics cards, boosting each other's performance and inflating valuation bubbles. Only the companies that provide data collection and sell graphics cards actually make substantial profits.
5. Temporary Profits: The Good Times for Data Collectors Are Limited
The current profit margin for data collection companies is a result of the industry's early stages. The industry is still far from the point where robots can be widely commercialized. There are three main hurdles to overcome: a lack of basic data, robots still in trial phases, and most companies losing money. Once enough data is collected or robots can generate data on their own, the current profit model based on low labor costs will come to an end. These data collection companies are essentially reaping the early benefits of the industry, but these good times will not last forever. The industry is still working to gather the necessary data and build the infrastructure for the future of robotics.