Momenta Races Toward Hong Kong IPO with Revenue Tripling to RMB 2.4 Billion Over Three Years
Alina Collins
Autonomous-driving company Momenta has filed for a Hong Kong IPO, with revenue growing from RMB 743 million in 2023 to RMB 2.41 billion in 2025 — a three-year CAGR above 80%. The bigger story: licensing revenue surged roughly 42×, the first scaled validation of a license-fee model in China's autonomous-driving market.
Revenue tripled in three years — what drove the growth?
Momenta's revenue rose from RMB 743 million (2023) to RMB 2.41 billion (2025), a three-year CAGR above 80%.
The structural shift came from licensing revenue: RMB 23 million in 2023 surged to RMB 968 million in 2025 — roughly 42× in three years.
This means → licensing income — fees automakers pay per vehicle to use Momenta's autonomous-driving system — flipped from a marginal line item to a core engine, jumping from ~3% to ~40% of total revenue.
Why is the licensing model seen as an industry milestone?
License-fee revenue works on a "sell one car, collect one fee" basis. The more vehicles shipped, the more revenue — with near-zero marginal cost.
In plain terms = it works like per-device software licensing: the development cost is already spent, so every additional car installed is close to pure profit.
Per CIC data, from March 2025 to February 2026, vehicles equipped with Momenta's system held a 65% share of China's third-party urban NOA — navigate on autopilot — supplier market by sales volume, ranking first.
What does "world model" mass production actually mean?
CEO Cao Xudong repositioned Momenta from "an autonomous-driving company" to "a builder of physical-AI foundation models". The flagship product is the R7 world model, already in mass production.
R7 draws on over 12 billion km of real driving data, distilled into more than 100 million training segments. The on-vehicle deployment is a distilled, lightweight version; its first production vehicle is the SAIC Volkswagen ID. ERA 9X.
This means → Momenta is no longer just selling a driver-assist feature. It is selling an AI foundation that models physical-world causality — a narrative shift that raises the ceiling on its valuation story.
How is the technical architecture structured — and what does each layer solve?
Layer 1: pre-trains physical laws and causal relationships from massive driving data — giving the system broad real-world exposure.
Layer 2: uses generative models to simulate extreme long-tail scenarios (wrong-way driving, irregular intersections) and runs closed-loop tests — stress-testing against rare dangers.
Layer 3: runs tens of millions of episodes of reinforcement learning in virtual environments — pushing the system from imitation to autonomous exploration.
In plain terms = the three layers stack up as: see a lot of roads first, then drill the hard cases, then let the system figure out optimal moves on its own.
How does the "flywheel" business logic work?
The core strategy is "one flywheel, two legs": over 900,000 L2 mass-production vehicles generate real-world data and commercial revenue, funding continuous iteration of the world model.
The improved model then deploys to L4 Robotaxis — fully driverless taxis — now operating in Shanghai, Suzhou, Munich, and Abu Dhabi.
This reflects Momenta's bet that L2 scale can fund L4 technology — production cars serve as "cash cow + data source," while Robotaxi serves as "tech validation + future narrative."
Is the physical-AI expansion path credible?
Whether autonomous-driving technology can migrate at low cost to robots and other physical-AI endpoints remains an open question in the industry.
Whether "predicting the next physical state" and "predicting the next token" are fundamentally the same class of problem is still debated in academia.
This means → Momenta's IPO sets a new valuation reference for the physical-AI track, but the real test for the market is: can licensing revenue sustain its high growth rate, and can the "physical-AI foundation" story convert into real, multi-endpoint revenue?
Content is for reference only, not financial advice.