Momenta Files for Hong Kong IPO: The First Physical AI Stock, World Model Goes into Mass Production

Momenta (Mengteng Intelligent Driving Global Limited) has passed the Hong Kong Stock Exchange listing hearing and is expected to start its IPO subscription in late June, potentially becoming the "first physical AI stock." The company has repositioned itself from autonomous driving to building a physical AI foundation model, with its R7 world model already in mass production, deployed in vehicles such as the SAIC Volkswagen ID. ERA 9X.

Key Financials and Business Performance

• Revenue: Grew from 743 million RMB in 2023 to 2.413 billion RMB in 2025, tripling in three years with a CAGR exceeding 80%.
• License Revenue: Surged from 23 million RMB to 968 million RMB, a 42-fold increase in three years, reflecting the high-margin License fee model.
• Market Share: From March 2025 to February 2026, Momenta held a 65% market share in China's third-party urban NOA supplier market, ranking first.
• Mass Production Scale: Over 900,000 production vehicles equipped with its system, cumulative delivery of over 100 models, and over 210 models under contract; fastest delivery of 100,000 units in under 40 days.
• Customers: Collaborations with 9 of the top 10 global automakers, including Mercedes-Benz, BMW, Audi, Toyota, Volkswagen, and BYD.
• R&D Investment: 1.869 billion RMB in 2025, accounting for 77.5% of revenue, with a cumulative 4.66 billion RMB over three years; 1,157 R&D personnel, representing 82% of employees.
• Cash Reserves: Exceeded 10 billion RMB by the end of 2025.

Technical Architecture: R7 World Model

Momenta's R7 world model adopts a three-layer architecture:

1. World Model Pre-training: Based on 100 million segments of "golden data" distilled from 12 billion kilometers of real driving mileage, compressing physical laws and causal relationships.
2. World Model Simulation: Generates models to deduce long-tail scenarios and conduct closed-loop testing, with efficiency far exceeding real-world road tests.
3. Reinforcement Learning: Optimizes decision-making autonomously through reward-punishment mechanisms in virtual interactions, enabling the model to surpass human performance in extreme scenarios.

The model is directly used for end-to-end foundation model pre-training, rather than merely as a simulation tool, creating a technological moat.

Strategic Positioning and Industry Background

Since its inception, Momenta has adhered to a "one flywheel, two legs" strategy: a data-driven flywheel, where L2 mass-produced assisted driving and L4 Robotaxi share the same software algorithms and world model. L2 production vehicles provide data and revenue to support model iteration; the iterated model is deployed in Robotaxi (already operational in Shanghai, Suzhou, Munich, Abu Dhabi, etc.), with extreme scenarios feeding back into training.

Physical AI is seen as the next wave after generative AI, with world models as the core foundation. NVIDIA CEO Jensen Huang has repeatedly emphasized this direction and launched the full-modal world model Cosmos 3. Momenta is the first to bring a world model to mass production, becoming a pioneer in this track.

Risks and Challenges

• There is no mature methodology for transferring autonomous driving technology to other physical AI terminals like robots.
• Whether "predicting the next physical state" is essentially the same as "predicting the next token" remains debated in academia.
• Intense competition: giants like Tesla and NVIDIA are also developing world models.