China now has its own AI circular financing scheme
Unitree Robotics achieved a ~$50 billion valuation in its Shanghai IPO with a 460% stock surge, becoming mainland China's first publicly traded humanoid robot maker Nearly 75% of Unitree's humanoid revenue comes from state-backed training centers that buy robots and sell training data back to manufacturers, creating a circular financing model Only 25-37.5% of collected training data is considered usable, as robots operate in non-real-world settings rather than actual deployment environments The
Analysis
TL;DR
- Unitree Robotics achieved a ~$50 billion valuation in its Shanghai IPO with a 460% stock surge, becoming mainland China's first publicly traded humanoid robot maker
- Nearly 75% of Unitree's humanoid revenue comes from state-backed training centers that buy robots and sell training data back to manufacturers, creating a circular financing model
- Only 25-37.5% of collected training data is considered usable, as robots operate in non-real-world settings rather than actual deployment environments
- The model mirrors criticisms of self-sustaining demand in the US AI industry (e.g., Nvidia's investments), though Chinese officials compare it to successful state-driven EV and solar panel dominance
- Unitree trades at 35.89x revenue versus ~20x for Hong Kong rivals, with analysts and early investors expressing skepticism about fundamental valuation support
Why It Matters
This circular financing model reveals how state-backed ecosystems can artificially inflate demand and valuations in emerging AI hardware sectors, raising questions about market sustainability and data quality. For AI practitioners and investors, it highlights the critical distinction between policy-driven revenue and genuine commercial demand—a pattern that could recur as humanoid robotics scales globally.
Technical Details
- Circular business model: Manufacturers sell robots to state-backed training centers (over 90 by June), where teleoperation "teaches" physical tasks; centers then resell the collected training data back to the same manufacturers
- Data quality concerns: A center manager reported only 2-3 of every 8 training hours are usable, as robots do not operate in real-world deployment conditions
- Revenue concentration: Unitree derived ~75% of humanoid revenue from education and research in the first nine months of 2025; Leju similarly saw 45% of flagship robot sales from training centers
- Valuation metrics: Unitree priced at 35.89x revenue vs. ~20x for Hong Kong competitors; IPO raised 6.1 billion yuan ($904 million) with ~20% allocated to strategic investors including Deepseek
- Cost structure: Training data for a five-minute robot dance can cost up to 1 million yuan ($148,000), illustrating the economic loop between data purchase and resale
Industry Insight
- Valuation risk: Circular demand models can create illusionary market sizing—investors should scrutinize whether revenue stems from end-user adoption or internal ecosystem transactions before committing capital
- Data utility gap: Training data from controlled/teleoperated environments may not transfer effectively to real-world deployment; companies should invest in diverse, in-situ data collection rather than relying solely on simulated or center-based datasets
- State-driven sector play: Beijing's strategy mirrors its EV and solar dominance, suggesting humanoid robotics will receive sustained policy support—but this also means geopolitical shifts or subsidy reductions could rapidly impact company fundamentals
Disclaimer: The above content is generated by AI and is for reference only.