Core Summary
Zuji Power is a leading company focused on humanoid robots (embodied intelligence), with financing backing from industry giants such as Alibaba, JD.com, and NIO, as well as global investors from the Middle East and Europe. Founder Zhang Wei presented several counter-conventional views at the 2026 World Artificial Intelligence Conference: Humanoid robots are in a period of exponential growth; while the hardware is not difficult to create, physical AI (the intelligence to understand the real world) is the bottleneck; the “brain” of a robot is a layered and collaborative system, not a single model; commercialization should follow the “universal body + app” approach, starting with specific use cases before iterating on the models; 2026 marks the beginning of the validation period for embodied intelligence, and the industry will see a multitude of innovations rather than just one dominant player.
Detailed Analysis
1. Behind the Financing: What’s the Logic Behind the Investment from Giants?
Zuji Power has completed multiple rounds of financing from 2023 to 2026, with investors including both industry capitals like Alibaba and JD.com, as well as financial/global funds such as Lei Shi Capital from the Middle East and IDG.
- **Industry capital values “scenario collaboration”: Alibaba’s first investment in embodied intelligence, and JD.com’s leadership in partnerships for retail/logistics applications—humanoid robots can help with tasks like grocery sorting in supermarkets and cargo handling in warehouses, directly addressing labor challenges for these companies; NIO and SAIC can provide their expertise in the automotive industry’s supply chain and manufacturing.
- **Global capital sees the “future market”: Investments from Middle Eastern funds (Lei Shi) and Italian consortia are driven by the high demand for humanoid robots in services (e.g., luxury hotels, desert rescue) in the Middle East and industrial applications in Europe; Zuji Power’s global strategy positions it well to capitalize on these opportunities.
- The advantage of the shareholder mix: Industry capital provides scenarios and resources, financial capital provides funding, and global capital opens up overseas markets—this combination enables Zuji Power to develop technology while quickly generating revenue.
2. Counter-Conventional Views: Are Humanoid Robots Really Easier to Build Than Cars?
Zhang Wei claims that “robot bodies are easier to build than cars,” which may seem counterintuitive, but there’s a rationale:
- Hardware is not complex: The number of components in humanoid robots is only one-tenth that of cars; technologies like joints and motors are relatively mature. Unlike semiconductors, which require nanoscale precision, or cars, which have complex safety standards, humanoid robots don’t face such challenges.
- **The real challenge is “physical AI”: Making a robot move is not the issue; making it function effectively (e.g., holding a cup without dropping it, climbing stairs smoothly) requires understanding physical laws in the real world. This data is scarce and costly to obtain through real-world testing (e.g., trying 100 times to learn how to hold a cup properly).
- Don’t be misled by their current limitations: Robots may seem primitive now, but they’re still in the early stages of development (similar to smartphones in 2015); using linear thinking (e.g., “if they’re clumsy now, they’ll always be clumsy”) will miss out on potential opportunities. By the time robots become truly useful, others might already dominate the market.
3. The Robot Brain: Not a Single Model, but a System of Operating Systems
Many think the robot brain is just a large model, but Zhang Wei believes it’s more like a layered and collaborative system:
- **Upper “thinking layer”: Similar to the prefrontal cortex, uses large language models (LLMs) to understand commands and plan tasks (e.g., delivering a package to Lily).
- **Middle “vision-motor layer”: Similar to the visual cortex, observes the environment and plans movements.
- **Lower “execution layer”: Similar to the cerebellum, controls joints and muscles to execute actions.
- The key is coordination: These layers must work together in milliseconds (e.g., adjusting balance immediately upon seeing a slippery surface) and optimize with hardware (joints/motors). This process is akin to rehabilitation for a disabled person, requiring coordinated brain-body training; Zuji Power calls this “integration of the cerebrum and cerebellum.”
- Current state: Robot brains are at the GPT-3 level, not perfect but capable of simple tasks (e.g., autonomous delivery), with significant room for improvement.
4. Commercialization: Universal Body + Apps; Don’t Wait for “Perfect” Solutions
How can humanoid robots make money? Zhang Wei’s approach is similar to that of smartphones:
- Universal body + apps: The body is the hardware, and apps are the specific functions (e.g., delivering packages, cleaning tables). Single functions are limited, but multiple functions can be added without changing the hardware. Once a certain threshold (e.g., 100 commonly used functions) is reached, more users will adopt the system—similar to how more apps on an Android phone attract more users.
- **Don’t follow the big model approach of “universal first, then application”: Big models can be generalized before finding specific use cases, but physical AI doesn’t work that way; data for real-world tasks (e.g., driving or egg-carrying) is too specialized. The correct path is to develop basic general capabilities and immediately apply them in practical scenarios, using data from those scenarios to improve the model in a virtuous cycle.
- Scenario focus: Zuji Power focuses on “human-centered” applications (e.g., homes, supermarkets) rather than industrial ones, where dedicated robots (e.g., robotic arms) are more efficient. Humanoid robots are better suited for human environments (e.g., climbing stairs, reaching high places).
5. Industry Trends: 2026 as a Year of Validation and Diverse Innovation
Zhang Wei predicts:
- **2026 is the year of PoC (proof-of-concept): More robots will transition from being demonstration tools to practical solutions (e.g., grocery sorting in supermarkets, household cleaning).
- Scalation next year: Mass production may begin in 2027.
- Ecosystems are more important than individual efforts: Zuji Power has open-sourced its training architecture for developers to train robot skills and is promoting an industry ecosystem. Embodied intelligence requires collaboration among component manufacturers, scenario providers, and developers; only through diverse development can the industry advance rapidly.
Final Summary
Humanoid robots are not just “future technology” but are evolving from toys to practical tools. Companies that invest now may reap the benefits of exponential growth in the coming years. Ordinary people don’t need to understand the technology, but they should realize that having a robot that can deliver packages or clean tables at home might be sooner than you think.