Gasgoo Munich-On July 8, 2026, ticker "06880" began trading on the Hong Kong Stock Exchange big screen. Momenta officially listed at 295.6 HKD per share, opening at 301 HKD—a 1.83% gain that gave the company a market capitalization of roughly 70 billion HKD. Turnover reached 364 million HKD. At the time of writing, following volatility, the stock turned positive again, trading at 303.4 HKD, up 2.64%.
Image source: Momenta
SAIC Volkswagen and Roewe extended congratulations immediately. Market participants repeatedly analyzed cornerstone investment notices from Mercedes-Benz and BYD, while the names of 14 top-tier institutions lined the back cover of the prospectus. It was, by all accounts, a demonstration of market confidence.
Image source: SAIC Volkswagen, Roewe
The stock's trajectory—surging at the open, dipping slightly, then recovering—reflects the capital market's complex attitude toward this new "Physical AI" entity. Chinese autonomous driving companies have listed in Hong Kong in recent years, but first-day performance has been weak. Trading below the issue price is the norm; only a few manage gains. The market is weary of the autonomous driving narrative.
That Momenta's share price held above its issue price amid volatility suggests the market takes the "Physical AI" label seriously—and prices it in.
Intense Competition in Autonomous Driving as "Physical AI" Becomes New Valuation Metric
Being an autonomous driving solution provider is no longer attractive to capital markets. Valuation logic is undergoing a significant shift. The sector has retreated sharply since late last year, and risk appetite has clearly decreased. In the A-share market, autonomous driving stocks are sluggish, with most players logging double-digit declines this year.
Why? Because the market realizes the business model has a natural ceiling.
First, automakers are increasingly developing technology in-house. Top OEMs want to control autonomous driving technology, squeezing third-party providers. Second, solutions are becoming commodities. With widespread adoption of highway and urban NOA, technical differentiation is blurring, and pricing power is eroding.
Crucially, patience for the "autonomous driving" tag is diminishing. From the hype of 2022 to the scrutiny of 2023, and the fatigue of 2024-2025, autonomous driving has shifted from a scarce capability to a standard feature—from a source of premium to a cost line item.
Momenta identified the crisis. So, in 2026, its narrative shifted.
Cao Xudong repositioned the company as a "builder of Physical AI foundation models"—creating a general world model that understands physical laws and simulates evolution. This was emphasized during the IPO roadshow: just as GPT provides the language base for digital AI, Momenta's world model provides the spatiotemporal base for physical AI.
This pivot elevates Momenta from a "component of the automotive supply chain" to the "AI infrastructure layer." The valuation logic shifted from a company selling algorithms to a platform defining the next computing paradigm. GIC, Fidelity, BlackRock, Oaktree, Franklin Templeton—these top long-term funds are betting on this narrative by backing an unprofitable company.
But narratives must eventually land. Whether the "Physical AI" story is valid depends on three things: data scale, model capability, and commercialization speed. All are essential.
A deeper question remains: Is the technical foundation of this "Physical AI" narrative genuinely different from the traditional "autonomous driving provider" narrative? Or is it just a more attractive marketing wrapper? Does the R7 world model truly "understand" the physical world, or is it still just complex pattern recognition? The capital market will ask these questions sooner or later.
Can the "Flywheel" Behind a 71.6% Gross Margin Keep Spinning?
Momenta's prospectus reveals a rare set of figures.
Revenue reached 2.413 billion yuan in 2025, up 82% year-over-year, with a three-year compound growth rate exceeding 80%. This speed is notable for a recent tech IPO. More striking is the gross margin—rising from 17.5% in 2023 to 49% in 2024, and hitting 71.6% in 2025.
The key to this margin increase lies in a qualitative shift in revenue structure.
According to Gasgoo, citing Momenta, corporate revenue splits into development fees and licensing service fees. High-priced, low-volume models rely on development fees; low-priced, high-volume models rely on licensing. Licensing revenue surged from 23 million yuan in 2023 to 968 million yuan in 2025—a 42-fold jump that lifted its share of total revenue from 3.1% to 40.1%. Since licensing has near-zero marginal cost, this is the core driver of the margin increase.
Regarding profit, adjusted net loss narrowed from 1.093 billion yuan in 2023 to 959 million in 2024, and further to 303 million in 2025. It is now about 300 million yuan from break-even. For a company with 2.4 billion yuan in annual revenue and 71% margins, that target is not far off.
However, there is a caveat to these positive figures: net losses on the books widened from 2.57 billion yuan in 2023 to 3.206 billion in 2024, and 3.458 billion in 2025. The cause is fair value changes in preferred shares and financial liabilities—non-cash items that do not directly impact operations. Yet these "paper losses" will test market patience, especially as Hong Kong becomes more cautious about unprofitable technology valuations.
Momenta's business model is a "data flywheel": mass production accumulates data, which trains models, which improves experience, attracting more OEMs, which yields more data. By February 2026, Momenta had secured design wins for 180 models, with 68 reaching SOP. Gasgoo reports that by July, it had equipped 1 million vehicles.
Image source: Momenta
The first 100,000 units in 2022 took 24 months; while the fastest 100,000 deliveries now take under 40 days.
The flywheel is accelerating. Nine of the world's top 10 automakers are partners.
Whether it continues depends on two factors. First, the impact of in-house development: as top OEMs catch up on urban NOA, will third-party premiums be squeezed? Second, the price war: will upstream cost pressure erode Momenta's pricing power and margins? The answers will emerge over the next 12 to 18 months.
However, Momenta has an edge rivals struggle to replicate: a "full-coverage strategy." Unlike peers focusing on mid-to-high-end models, Momenta spans the entire price range from 100,000 to 1 million yuan. High-end models bring development fees; mass-market models bring recurring licensing revenue—a combination that secures a healthy margin structure.
Is Physical AI's "GPT Moment" Real, or Just Marketing Spin?
In April, Momenta released the R7 reinforcement learning world model, announcing its first mass production launch.
Cao Xudong claims this marks a leap from "seeing the world" to "understanding the world"—from perceptual intelligence to cognitive intelligence. The rhetoric invites comparisons to ChatGPT's paradigm shift in digital AI. But R7's technical reality requires objective analysis.
R7's three-layer architecture warrants analysis.
First, world model pre-training. Using over 12 billion km of real-world driving and more than 100 million refined "golden clips," it compresses physical common sense and causality into the foundation model. The breakthrough is teaching AI concepts like "inertia," "collision consequences," or "a pedestrian looking back implies lane change intent"—beyond simple memory and rule matching. Data quality exceeds quantity here; those 100 million clips are long-tail scenarios distilled from millions of cars, forming a data moat.
Second, world model simulation. Generative models simulate the environment for closed-loop testing, allowing evaluation of rare long-tail scenarios. Efficiency is tens of thousands of times higher than real-world testing. Momenta's advantage lies in aligning simulation with real-world data, offering a reliable benchmark compared to pure rendering.
Third, reinforcement learning within the world model. Through reward mechanisms, the model explores and tests errors in a virtual world. The system moves from "imitation learning" to "imagination and exploration," experiencing millions of simulations to handle extreme cases rarely seen on real roads.
The core difference is this: the industry typically uses world models as simulation tools to generate data for a main model, whereas Momenta applies it directly to "end-to-end foundation model pre-training." The former is a mock exam; the latter reshapes the student's cognitive structure.
Cao Xudong's thesis: once AI approaches human level, it surpasses us rapidly. "It might take ten or twenty years to climb the slope, but surpassing humans happens in a year or two."
But this narrative faces a fundamental challenge: how to prove that "understanding physics" beats "memorizing scenes" in real driving? R7 just launched; scaling verification takes time. If the experience falls short, "Physical AI" risks becoming marketing spin.
Another risk: the cycle from R7 release to mass production was extremely short. Was engineering verification sufficient? While reinforcement learning excels in simulation, has the sim-to-real gap been bridged? Only large-scale real-world operation can resolve these questions.
Momenta's timeline is clear: complete L4 Robotaxi scale verification in 2026, officially launch L4 Robotruck operations in 2027. These milestones will be the proving grounds for "Physical AI."
Notably, the world model holds strategic value often underestimated by capital markets: cross-scenario transferability. Once it masters general physical laws, this "physical brain" can generalize to Robotaxis, Robovans, Robotrucks, and even humanoid robots. CIC predicts the global Robotaxi market will reach $81.8 billion by 2030, Robovan $85 billion, and Robotruck $33 billion—totaling nearly $200 billion. This is the deep logic supporting Momenta's high valuation.
69.5 Billion HKD: Golden Start or Valuation Bubble?
Back to the core question: 69.5 billion HKD—is it expensive?
Based on 2025 revenue, Momenta trades at roughly 24 times sales—far pricier than peers. The market is paying a scarcity premium for the "first Physical AI stock."
Whether that premium holds depends on three judgments.
The bull case is solid. Nine of the world's top ten automakers are customers, and seven invested strategically pre-IPO: SAIC, GM, Mercedes-Benz, Toyota, BYD, Hyundai, and Chery. This "automotive circle" carries more weight than any rival's. Mercedes-Benz, for instance, invested in 2017 and won mass production by late 2025—an eight-year integration creating an unbreachable moat.
Image source: Momenta
The cornerstone investor lineup is also telling. Mercedes-Benz and BYD are both existing shareholders and new cornerstone investors. Mercedes represents the global auto industry's demand for safety and performance; BYD represents the new energy market's demand for scale and delivery. Just as Microsoft backed OpenAI and Google backed Anthropic, automakers are backing Momenta to lock in core capabilities for the Physical AI era.
But the pressure cannot be ignored. The industry is consolidating rapidly. Cao predicts strong scale effects and first-mover advantages, leaving only 2-3 winners in China and 3-4 globally. Can Momenta hold the lead? Competitors like Huawei ADS, Deeproute, and QCraft are accelerating. Urban NOA penetration was about 11% in 2025 but is expected to hit 62% by 2030. The incremental market is huge, but the battle is far from over.
Financial pressure is also mounting. While gross margins are high at 71.6%, net losses are widening. In a risk-off cycle, unprofitable high-valuation technology stocks are the first to suffer. Moreover, a 24x P/S prices in growth expectations for 2026-2027. If revenue growth slips below the 50-70% range, the risk of a valuation bubble is real.
There is also an overlooked macro variable: the valuation anchor for the entire autonomous driving sector is drifting. As the market tires of the "autonomous driving" tag, Momenta must constantly prove via the "Physical AI" narrative that it belongs to a new, viable sector. It's an opportunity and a risk—the chance to be the only one to escape the old narrative, or the risk that Physical AI is just "old wine in new bottles," triggering a sharp mean reversion.
The deeper question: What will Momenta ultimately become? A "Tesla that doesn't make cars," using a single technology stack from L2 to L4 to become a general platform? Or an "Anthropic of the physical world," achieving commercial closure in vertical scenarios before generalizing? Or merely a hyped component in a valuation game?
Conclusion:
Momenta's stock held above its issue price on debut—the market offered a warm welcome but maintained scrutiny. Whether this "Physical AI" story can sustain a 70 billion HKD valuation will not be found in today's candlestick chart. It lies in the data from every road test, every new design win, and every quarterly report.
The most challenging and honest aspect of Physical AI is this: you can narrate "understanding the physical world" a thousand times, but once the car drives out—whether it crashes, whether it is smooth, whether users buy it—you cannot fool the market. As Cao Xudong put it: "It might take ten or twenty years to climb the slope, but surpassing humans happens in a year or two." Momenta has climbed for ten years; now it's time to let the product speak.
