Summary of Key Points
In the past six months, 46 billion yuan has flowed into the field of embodied intelligence (robots that can perceive and interact like humans). Capital has concentrated in leading companies, with industrial capital (such as Baidu and ByteDance) and local state-owned assets playing a significant role in financing. The industry is hailed as the "year of mass production," but robots that can actually perform tasks reliably have yet to appear in large numbers. Technically, the "brains" of these robots (AI models) are not mature due to a lack of real-world interaction data. Commercially, raising a lot of money does not necessarily equate to success; for example, a company once valued at 20 billion yuan, Dada, only sold 1.4 million units in sales. While local governments and state-owned entities are entering the market, there are also risks similar to those seen in the early stages of the photovoltaic industry. Whether this will turn into a trillion-dollar opportunity or just impressive demos depends on whether robots can solve real-world problems in practical applications.
I. Capital Frenzy: Who Got the Money?
The 46 billion yuan invested over the past six months is equivalent to moving hundreds of thousands of yuan back and forth between Beijing and Shanghai. However, the money was not distributed evenly: the top five companies (such as Qianxun Intelligence and Xiwang) received 37% (17.1 billion yuan), and the first twenty companies accounted for 70% (33 billion yuan), with the remaining more than 200 companies getting only a fraction of that amount. Qianxun Intelligence, in particular, raised 4.5 billion yuan in four rounds within four months.
Who is funding this? Traditional venture capitalists (such as Hillhouse and Sequoia) are still active, but the main players in large-scale financings (over 1 billion yuan) are now industrial capital and local state-owned assets. Baidu has invested in Zhi Ping Fang and the Beijing Humanoid Robot Innovation Center; Meituan and Didi have invested in Diguagua Robotics; SAIC invested in four companies within half a year; local state-owned assets accounted for 42% of these large transactions. The logic behind local governments is straightforward: provide funding → require companies to set up factories locally → use those factories as their first customers (for instance, Shenzhen has established the "Embodied Intelligence Port" to attract companies like Tencent).
II. The Dada Warning: Much Financing Does Not Equal Longevity
The industry claims it's the "year of mass production," and Yushu sold 5,500 humanoid robots last year, with revenue increasing from 159 million yuan to 1.699 billion yuan; Zhiyuan launched its first robot in March. However, these numbers hide issues: Dada, which raised 54 billion yuan and was valued at over 20 billion yuan, only sold 1.4 million units in the first seven months of 2025, resulting in a net loss of 84.25 million yuan.
This shows that raising money does not guarantee profitability. A common problem is that robots look good but are not practical; they may be able to perform simple tasks like tightening screws or folding clothes during demonstrations, but few can handle complex factory processes. Customers find them ineffective in different scenarios and are therefore unwilling to pay.
III. Technical Hurdles: The Robots' "Brains" Are Not Yet Advanced
The biggest issue with robots is not their lack of mobility but the inadequacy of their "intelligence." The core bottleneck is the scarcity of real-world interaction data: there are only 500,000 hours of usable data for robot training globally, compared to the 20,000 times more data used to train large language models like ChatGPT. Robots cannot learn from online data; they need to test in real environments repeatedly. Companies like Xinghai Tu have invested heavily in a "million-hour data collection project," while Qianxun has set up 300,000 data collection points and Ant Lingbo has analyzed 20,000 hours of data for training models. However, the results are disappointing: an algorithm expert said that spending tens of millions on collecting 100,000 hours of data only improved model performance by 5%, and skills learned in one factory may not apply in another.
Model maturity is also a significant issue. Experts compare the ultimate capability of robots to a score of 100: industrial robotic arms currently score around 50, bipedal humanoid robots about 15, dexterous hands around 5, and accompanying AI systems around 3. Moreover, there is no clear way to objectively evaluate these models—industry benchmarks often focus on demonstrations, but real users need to see if robots can perform tasks effectively in new contexts.
IV. The Entry of State-Owned Entities: A Boost or a Concern?
Local governments and state-owned entities are taking action to support the industry: Beijing is focusing on AI and robot hardware, Guangdong on components and joints, and Jiangsu, Zhejiang, and Shanghai on industrial applications (such as spraying and cleaning). Hangzhou has issued the first national regulation for embodied intelligence, while Shanghai aims to have 100,000 robots in factories by the end of the "15th Five-Year Plan." The Ministry of Industry and Information Technology has launched a "real-scenario training project" to deploy robots in practical settings by the end of the year.
However, these efforts come with risks: using state-owned funds means accepting geographical restrictions (such as local factory establishment), performance-based agreements, and exit restrictions. The photovoltaic industry's experience shows that excessive capacity can lead to losses; if the same happens with embodied intelligence, the consequences could be severe, given the current level of state-owned investment.
V. The Future: A Bubble or a Real Opportunity?
Optimists argue that industries like railways, the internet, and renewable energy also went through initial capital-intensive phases before becoming successful. Pessimists point out that only 13 billion yuan was invested in seed and angel rounds in the first half of the year, and young entrepreneurs without major company backing or academic credentials struggle to attract investment.
This is a common challenge for emerging industries: many companies will fail, but those that survive will drive the industry forward. The answer lies not in financing news or rankings but in the robots themselves—can they perform their tasks effectively? Will customers place additional orders? Foreign engineers use robots because their actions are useful, not just because they look impressive?
In summary: Embodied intelligence is a promising field, but it is still in the early stages of development and far from generating significant profits. While capital inflow is positive, we should focus on whether robots can solve real-world problems, not just on impressive demos.