Summary of Key Points
Momenta went public on the Hong Kong Stock Exchange on July 8th, becoming the “first stock in the field of physical AI,” with its shares opening more than 6% higher and a market value exceeding HK$70 billion. Founded by Cao Xudong, the company’s core strategy is based on the concept of “one flywheel with two legs”: using mass-produced L2-level assisted driving to accumulate vast amounts of data, which in turn feeds into more advanced L4-level autonomous driving systems. To date, Momenta has produced over 1 million vehicles, with plans to reach 2 million by the end of the year and over 10 million by 2028. The company positions itself as a leader in “physical AI,” distinct from digital AI technologies like ChatGPT, focusing on intelligent decision-making in the real physical world (such as autonomous driving and future home robots). It introduces the “World Model” as its core technology, with the goal of bringing “AI assistants, doctors, and teachers” to every household. Cao Xudong’s entrepreneurial journey has been full of unconventional choices: quitting Tsinghua University, working at Microsoft, and then founding Momenta, while adhering to the L2-level autonomous driving approach. His management style is influenced by mountaineering principles, emphasizing steady progress and team consensus.
Detailed Analysis
1. What exactly is “physical AI,” and how does it differ from ChatGPT?
AI can be divided into two categories: digital AI (like ChatGPT), which processes text, images, and videos using large amounts of digital data; and physical AI, which deals with real-world problems (such as determining whether a vehicle in front will brake or how a robot should handle objects). Momenta claims to be the first stock in physical AI because it focuses on enabling AI to understand and interact with the real world, rather than remaining at the digital level.
The “GPT moment” for physical AI refers to the development of a “world model”—similar to how ChatGPT compresses knowledge from the digital world into a model, a world model compresses the laws of the physical world (e.g., object movement, traffic rules) to allow AI to predict future events. This is more advanced than mere perception; it involves understanding the reasons behind phenomena and predicting their outcomes.
2. Why did Cao Xudong choose L2-level assisted driving instead of fully autonomous driving?
When Cao Xudong started his company in 2016, the industry was focused on L4-level fully autonomous taxis (e.g., Waymo), but he chose to focus on L2-level assisted driving. The reason is simple: without massive amounts of real-world data, L4-level technology cannot be achieved. His approach relies on the “flywheel effect”: the more L2 vehicles are sold (over 1 million so far), the more data is collected; the more data there is, the smarter the AI models become, and the better the L4 technology improves. This, in turn, makes L2-level systems more useful, attracting more car manufacturers to collaborate. In 2023, Momenta’s revenue was HK$743 million, expected to grow to HK$2.413 billion by 2025, with a gross margin increasing from 17.5% to 71.6%, and license service revenue increasing by 41 times. This success is due to the data barrier created by these millions of vehicles covering billions of kilometers each year, which covers almost all extreme scenarios (e.g., rainy weather, construction areas).
3. How did Cao Xudong transform Momenta from a loose research institute into a capable team?
Initially, Momenta resembled a loosely structured research organization where individuals worked independently without coordinated product development. In 2018, Cao Xudong decided to shift the company’s culture from a research-oriented to an entrepreneurial one. This transition was painful, with many employees leaving. He learned that most people only believe what they can see, so he focused on achieving small victories (e.g., completing the first mass-produced project, making technological breakthroughs) to build team cohesion. He also encouraged executives to read books like “The Longest Day” and Ren Zhengfei’s speeches to create a shared vision. Additionally, he implemented the “mainline razor” principle, which means using the simplest solutions to address the most common problems. For example, he merged parking and driving systems into one, despite the risks involved. This decision has positioned Momenta at the forefront in centralized architecture.
4. Future plans: Home AI assistants and the next steps for physical AI
Cao Xudong believes that autonomous driving is just the first step in physical AI; it marks the beginning of “embodied intelligence” (where vehicles become the “body” of AI). The next phase involves developing home service robots, such as AI assistants for household chores, health monitoring, and learning support. He chose the home domain because it leverages existing autonomous driving technologies, data collection methods, and world models, and has significant social value—bringing AI into households to improve daily life. His goal is to create a “GPT moment” for physical AI, just as ChatGPT transformed the digital world.
5. The CEO with a mountaineering background: A steady approach to entrepreneurship
Cao Xudong and many of his core team members come from Tsinghua University’s mountain climbing team, having climbed Mount Qier (6,168 meters). He compares entrepreneurship to climbing the Himalayas: it’s not about reaching the summit quickly (like in Alpine-style climbing) but about building bases and camps step by step. This approach is reflected in his decisions, such as sticking to the L2-level autonomous driving strategy and taking a gradual approach to organizational changes. He emphasizes positive feedback, believing that strong beliefs are built on facts and progress.
Momenta’s listing is not the end but a new milestone for the company and the field of physical AI. It demonstrates that a strategic focus on data accumulation through steady progress is effective and shows the potential for AI to extend beyond the digital world into the real world. Cao Xudong’s story shows that unconventional paths are possible as long as there is clear logic and persistence. Will physical AI change our lives in the same way ChatGPT has? We’ll have to wait and see.