By Da Cheung
Chinese autonomous driving software provider Momenta has moved one step closer to its highly anticipated IPO after passing its listing hearing at the Hong Kong Stock Exchange on Tuesday.
Analysts who follow the company have given it a pre-IPO valuation of over 100 billion yuan ($13.8 billion). Founded in 2016 by former SenseTime AI researcher Cao Xudong, Momenta has seen strong growth over the past three years. Revenue soared from 743 million yuan in 2023 to 2.41 billion yuan in 2025 largely driven by high-margin software licensing fees, which jumped from 23 million yuan to 968 million yuan over the same period, according to the company’s prospectus. By the end of 2025, it had amassed a cash reserve of more than 10.9 billion yuan.
However, as Momenta heads toward its public debut, the company is tweaking its narrative. In an era where investors are captivated by AI, simply being an “autonomous driving” company is no longer enough to woo the market.
A new narrative: physical AI
To stand out, Momenta is now positioning itself as the first physical AI stock, according to IPO Early News.
Unlike large language models that process text, physical AI relies on multi-modal inputs like video, laser radar, and ultrasonic sensors to understand and predict real-world physics. The company says its core technology, the R7 world model, uses these sensory inputs to understand and simulate driving environments and train itself through reinforcement learning. Momenta claims this three-tier architecture — world model pre-training, simulation, and reinforcement learning — enables the system to autonomously learn optimal decision-making without human intervention, creating what it calls a “GPT moment” for the physical world. In addition to current autonomous driving systems for passenger cars, the company also hopes the new world model will be installed on robotaxis and robovans.
Other Chinese firms are following suit. At the 2026 Beijing Auto Show, local competitors including Alibaba-backed DeepRoute, Zhuoyu Technology (a spinoff of dronemaker DJI), and Pony.ai (2026.HK) all pivoted their narratives toward physical AI. DeepRoute even dropped the word “driving” from its official Chinese branding and hired a former researcher from DeepSeek to lead its AI efforts. These companies argue that their physical base models will eventually enable machines not just to drive, but to operate heavy factory equipment or dance.
The data flywheel and Tier 1.5 strategy
Momenta’s business model hinges on what it calls a “one flywheel, two legs” strategy. It operates as a “Tier 1.5” supplier — a unique industry position where a company provides flexible software algorithms and decoupled solutions directly to automakers. This positions it not as a traditional hardware-centric supplier, but more as a strategic partner that collaborates with an automaker’s in-house development team.
The company provides mass-market vehicles with advanced driver-assistance systems, specifically urban NOA (navigate on autopilot). NOA is a feature that allows a car to automatically navigate complex city streets and is widely treated as a bridge between Level 2 and Level 3 autonomous driving. The data collected from these passenger cars is then used to train and refine algorithms for fully autonomous Robotaxis, which in turn improves the mass-market software.
This approach has secured Momenta partnerships with major global automakers, including Toyota, General Motors, and Mercedes-Benz. The company says its systems are now installed in over 900,000 vehicles. According to industry data cited by analyst Turen Chips, Momenta and Huawei collectively control over 80% of the third-party urban NOA market in China.
The Tesla challenge
As Momenta prepares for its IPO, the competitive landscape in China is intensifying, particularly with Tesla pushing to introduce its Full Self-Driving (FSD) system to the Chinese market.
While FSD’s entry raises the benchmark for driving software in China, industry analysts argue it may not be an existential threat to local leaders like Momenta or chipmaker Horizon Robotics (9660.HK). Strict Chinese data compliance laws require all domestic driving data to be stored locally. This severs Tesla from its global data training loop, meaning it must adapt its algorithms relying solely on local road videos rather than its vast international database.
Furthermore, Tesla’s FSD subscription in China is priced at about $140 per month. In a highly price-sensitive market where local competitors often bundle their software for a fraction of the price or include it as a standard feature, FSD is likely to remain a niche product.
Momenta’s localized data flywheel — processing data generated entirely within China’s borders — might give it a structural advantage over foreign rivals. To hedge against domestic competition and seek global growth, Momenta also formed a strategic partnership with Uber in April to expand its autonomous driving technology internationally.
While Momenta touts its rapid deployment and claims to have accumulated over 12 billion kilometers of real-world driving data, the true test will be whether its physical AI narrative can maintain momentum when it arrives on the public market amidst shifting investor sentiment and cutthroat price wars in the Chinese auto industry.
Sources:
