Key takeawayInsufficient robot generalization capabilities are driving demand for high-quality training data, while Yili’s revenue recovery and improved gross margin less selling expense ratio coincide with continued pressure on margins from a higher tax burden.
Why does robot training require more collected data?
- Wen Ji of Zheshang Securities believes a key obstacle to the widespread adoption of robots is their brains’ insufficient ability to generalize, with inadequate quantity and quality of training data among the underlying causes. At the start of 2026, the volume of real-world physical interaction data available for compliant use globally still fell 20% short of the scale needed to train a single general-purpose embodied intelligence foundation model.
- Data collection records real-world scenarios using sensors, cameras and human control, providing material for robot training. Existing approaches include teleoperation, EGO, UMI and simulation; EGO is relatively inexpensive and is mainly used to pretrain foundation models, while simulated data has the potential for low-cost, large-scale production.
- Wen Ji is optimistic about rapid growth in the scale of data collection and companies with confirmed orders, including Leyard, Lingyun Optical and Speechocean; the industry’s development also faces constraints involving data standards, quality, price competition and payment collection.
Where is the improvement in Yili’s revenue coming from?
- In the first three quarters of 2026, Yili’s revenue from liquid milk, milk powder and dairy products, and cold drinks was 55.72 billion yuan, 24.82 billion yuan and 10.12 billion yuan, respectively, up 1.4%, 2.3% and 7.3% year over year. Third-quarter liquid milk revenue rose 1.7% year over year, returning to positive growth; milk powder and dairy products grew 3.8%.
- New liquid milk products and emerging channels supported the recovery: membership stores, snack stores and on-demand retail drove 49.7% growth in direct sales channel revenue. In milk powder, Ausnutria’s third-quarter revenue decline narrowed from the second quarter, reducing its drag on the overall business; Huatai Securities expects Jinlingguan to potentially benefit from increasing industry concentration.
Why did net margin fall despite the improvement in gross margin less selling expense ratio?
- In the first three quarters of 2026, Yili’s gross margin rose 0.5 percentage points year over year to 36.0%, while its selling expense ratio fell 0.4 percentage points to 17.9%; the difference between the two widened by 0.9 percentage points to 18.1%. However, a higher effective tax rate caused the net margin attributable to the parent company to fall 2.1 percentage points year over year to 9.4%.
- The benefit from raw milk costs was smaller in the third quarter than in the second, while an improved product mix made a positive contribution; some products were reclassified as deeply processed products, increasing the tax base.
- Huatai Securities expects the timing of stocking for the Spring Festival to potentially support stronger fourth-quarter revenue performance than in the third quarter. It forecasts Yili’s net profit attributable to the parent company at 10.043 billion yuan, 13.088 billion yuan and 13.578 billion yuan for 2026 through 2028, respectively, and applies a 2027 price-to-earnings multiple of 17 times, corresponding to a target price of 35.19 yuan. The company has pledged that its 2026 dividend payout will be no less than the 7.7 billion yuan paid in 2024.
