Embodied AI's Real Problem: Lack of Genuine Orders and the Scene-First Path
A founder who recently took his company public on the Hong Kong stock exchange publicly questioned the prevalent "assembled entrepreneurship" in the industry, pointing out that such models rely on related-party transactions with data collection centers and leasing companies to create fake and unsustainable revenue, lack product-market fit, and rush to IPO within two to three years.
Almost simultaneously, media reported that regulators plan to tighten IPO reviews for humanoid robot companies, focusing on sustainable revenue, loss improvement, and genuine innovation.
The core debate is how embodied intelligence can prove its valuation credibility in the second half; the key lies in entering real scenarios, staying close to industry needs, and delivering verifiable orders.
Real orders must pass three tests. First, the order's quality: customers choose to buy after fully comparing competitors. Second, whether it solves real demand: capital markets focus more on customers' willingness to repurchase. Third, engineering capability from technology to scale delivery, requiring supply chain management, quality control, and cost control.
These three tests correspond to thresholds at different stages. Quality corresponds to product competitiveness, repurchase to demand authenticity, and scale delivery to engineering capability. Data collection centers and related-party transactions can create seemingly impressive contracts but cannot pass any of these three tests.
A third approach exists in the industry: grow capabilities from real scenarios. Jiyi Intelligent, founded by Jieyi Technology with 15 years in auto parts manufacturing, delivered products in just 8 months, launching industrial inspection robot Zhijianjia and commercial service robot Xiaomu, and released its self-developed Z-1 embodied large model.
The company started with automotive high-voltage wiring harness inspection, finding a real manufacturing pain point. This process has long relied on manual labor, with high miss rates, costly errors, difficult standardization, and high turnover. In tests, the Zhijianjia robot improved detection efficiency by 200%. Service robot Xiaomu has been deployed in chain pharmacies, providing 24-hour unmanned night pharmacy services.
The scene-first logic differs from assembled entrepreneurship in its starting point. Assembled startups begin with financing and valuation, first creating high revenue figures, then seeking scenarios. Jiyi identified a specific pain point with a clear payer from Jieyi's existing auto parts business and solved it with robots. In auto wiring harness inspection, customers already pay for manual inspection; if the robot is more stable and cheaper, they have reason to switch. This logic does not rely on related-party transactions or government orders but on real production line performance.
Pharmacy night unmanned service is another similar scenario: 24-hour pharmacies need night staff, and robots can replace this labor, making the economics work. Embodied AI financing over the past two years has far exceeded industry scale, and the scarcity of real orders is the core reason for this gap. Whether the scene-first path can scale depends on whether Jiyi can deepen its auto inspection and pharmacy services and replicate them to more similar scenarios.