At the end of August, Horizon Robotics (09660.HK) released its most closely watched interim report since going public: in the first half of the year, operating revenue reached 2.055 billion CNY, a year-on-year increase of 32.9%. Revenue from licensing and services exceeded chip sales revenue for the first time. The Journey series chips shipped 2.218 million sets, and its market share in the basic ADAS chip market for domestic brands surpassed 50% for the first time.
Almost at the same time, Volkswagen's first sedan in China supporting city NOA, the ID.ERA 5S, went on sale at a limited-time price of 114,900 CNY, equipped with Horizon Robotics' Journey 6 series chips and an end-to-end large model.
Additionally, on August 25, the draft revision of the Road Traffic Safety Law was submitted for deliberation by the Standing Committee of the National People's Congress for the first time, adding a special chapter on autonomous vehicles.
News from the three fronts of policy, chips, and complete vehicles converged in early September, pointing to the same shift: city NOA, once considered exclusive to vehicles priced above 300,000 CNY, is penetrating the 100,000 CNY and 150,000 CNY vehicle markets at an unprecedented pace. The democratization of intelligent driving is no longer just a slogan at launch events, but a reality jointly supported by regulations, costs, and production capacity.
01. Policy: The Line of Responsibility is Drawn at Automakers for the First Time
What first materialized in the democratization of intelligent driving was not price, but rules.
On July 2 this year, the mandatory national standard "Intelligent Connected Vehicles - Safety Requirements for Combined Driving Assistance Systems" (GB 47955-2026) was approved, scheduled to take effect on January 1, 2027. It sets the safety baseline for Level 2 combined driving assistance, and requirements such as driver monitoring and hands-off detection will directly raise the threshold for entry-level solutions. On July 30, the "Intelligent Connected Vehicles - Safety Requirements for Autonomous Driving Systems" (GB 44721-2026) for Level 3 and Level 4 was released, planned to be implemented on July 1, 2027. This is the country's first mandatory national standard for advanced autonomous driving, which makes clear requirements for perception redundancy and the minimum forward detection distance at a speed of 120 km/h. On August 25, the draft revision of the Road Traffic Safety Law added a special chapter on specific regulations for autonomous vehicles, stipulating that traffic violations occurring when autonomous functions are activated will be handled by the production and import enterprises. The draft was published for public comment on August 28.
Looking further back, at the end of 2023, the Ministry of Industry and Information Technology and three other ministries launched the pilot program for the access and on-road traffic of intelligent connected vehicles. On December 15, 2025, the first batch of Level 3 conditional autonomous vehicle models was approved for pilot operation in designated areas in Beijing and Chongqing. Thus, China's institutional framework for autonomous driving has completed three steps: local pilots, national standards, and national legislation. Cui Dongshu, Secretary-General of the China Passenger Car Association (CPCA), commented that this revision means the industry is moving from "patch-style corrections" to "chapter-style reconstruction." For the first time, the transfer of the responsible entity from drivers to automakers has obtained a basis in higher-level law, and automakers can no longer rely on rhetorical packaging to evade responsibility.
Rules and penetration rates have arrived almost simultaneously. The Ministry of Industry and Information Technology disclosed that since the beginning of this year, the penetration rate of passenger vehicles with Level 2 combined driving assistance functions has reached 70.5%, and the penetration rate of Navigate on Autopilot (NOA) has reached 34.2%. Seven out of every ten new cars are equipped with advanced driver assistance systems (ADAS). Once the rules lag behind, the risks will rebound on the industry itself. After the line of responsibility is clearly drawn, the biggest uncertainty on the path to volume scaling is actually eliminated: automakers dare to push functions on a large scale, users dare to actually use them, and the supply chain dares to arrange production capacity over a longer cycle.
Rules are also influencing technology selection. When interpreting GB 44721-2026, some institutions pointed out that the standard makes clear requirements for perception redundancy. Multi-sensor fusion solutions are more likely to meet compliance requirements, while the pure vision route has to bear higher verification costs. This means that regulations are not just a safety baseline but will also influence the hardware configuration structure in the next two years. The process of LiDAR moving from an optional accessory to standard equipment may be faster than pure cost logic suggests. It should be noted that the national standard does not specify the mandatory use of specific sensors such as LiDAR, and the competent department also emphasized in the interpretation the retention of multiple technology routes. The pure vision solution is not without room for development.
02. Concept: City NOA and Its Triple Leap Behind
The full name of NOA is Navigate on Autopilot, which refers to navigation pilot assistance. Highway NOA addresses car following, lane changing, and entering/exiting ramps on closed roads; city NOA has to face open scenarios such as traffic light intersections, unprotected left turns, and pedestrians and non-motorized vehicles crossing. It also needs to make judgments close to those of an experienced driver in traffic with continuous lane changes and cut-ins.
According to the national standard "Classification of Driving Automation", it still belongs to the L2 category. The driver must monitor the entire process and take over at any time. However, the complexity of the working conditions has already reached the ceiling of L2, and it is usually regarded as the entry point to L3. This is also why every time regulations mention full responsibility for L2 and automaker responsibility for L3, city NOA stands at the center of the topic.
The premise for the democratization of intelligent driving to hold true is that the technology route has completed three switches in three years.
Specifically, early solutions relied on high-definition maps and a large number of hand-written rules, resulting in high costs and slow city expansion. The introduction of end-to-end large models allows the system to learn how to drive directly from massive driving data, no longer relying on engineers to write rules one by one. Subsequently, the industry shifted to a mapless route represented by heavy perception and light maps. Coupled with new architectures such as VLA (Vision-Language-Action) and world models, the cycle for opening new cities has been compressed from months to days, and the collection and maintenance costs of high-definition maps have been almost saved. VLA can be simply understood as a model that allows the system to make judgments by looking at images like a human, while a world model enables it to predict what might happen on the road in the next few seconds.
What technical simplification brings is a synchronous drop in hardware costs: the domain controller for city NOA, which is the computing unit on the car responsible for aggregating and processing camera and radar data, has seen its cost drop from about 8,000 CNY to the 2,000 CNY level; LiDAR has dropped from over 100,000 CNY in the early years to around 1,000 CNY. When a medium-computing-power chip can run end-to-end in urban areas and sensors no longer cost tens of thousands of CNY, advanced intelligent driving has the economic viability to become a mass-market configuration for the first time.
03. Supply: Who Drove the Prices Down
For city NOA to enter the mass-market price range, the first to take action were the OEMs.
Entering 2026, advanced intelligent driving has almost simultaneously become a standard move for various automakers to boost volume. BYD has rolled out its God's Eye solution to high-volume models, with LiDAR following into new cars in the 100,000+ CNY price range; Chery's Falcon solution has installed city navigation pilot into fuel vehicles under 150,000 CNY, with the Arrizo 8 PRO being the first model to try it at this price point; Volkswagen, through its joint venture CARIZON with the Chinese team, developed a full-scenario advanced solution based on the Horizon Journey 6 series, covering seven models of the three joint ventures in China, which will be launched in the third quarter. The most direct statement is undoubtedly the Volkswagen ID.ERA 5S, which went on sale at a limited-time price of 114,900 CNY, effectively bringing city NOA, which was previously only available in the 300,000 CNY class, down to the 110,000 CNY tier.
Unlike the previous practice of being exclusive to top-trim models and only appearing on new cars, this round of democratization adds two more paths. One is activating existing inventory: Li Auto enabled city NOA for the first time on sold models equipped with the AD Pro platform through a version update, adopting a combination of one Journey 6M chip and the QCraft solution, with costs significantly lower than previous solutions; Changan Qiyuan also plans to push the city navigation pilot function to users via OTA in mid-October. The other is to continue exploring downwards: brands like Leapmotor have achieved a breakthrough from zero to one in the assembly rate of city NOA in the sub-100,000 CNY market.
The confidence of automakers to drive prices down to this level comes from the localization of upstream chips. The Horizon Journey series shipped 2.218 million sets in the first half of the year, a year-on-year increase of 12.1%. Among them, the Journey 6B has achieved global first mass production on GAC Toyota models, cumulatively winning mass production projects from over 25 mainstream automakers globally, with an estimated full-cycle demand of over 20 million sets. The combined orders for products developed by Bosch and Zhijia Xincheng based on the Journey 6B exceed 10 million sets, with mass production starting as early as the third quarter. Share data shows how fast the changes are: in the city NOA chip market, Horizon's share increased from 17.9% in 2025 to 22.8%, and its ranking rose from third to second in the industry; in the basic ADAS chip market, its share in the overall market was less than 5% in 2022; by the first half of this year, its share in the domestic brand market alone had surpassed half. The revenue composition is also changing. In the first half of the year, revenue from licensing and services increased by 52.7% year-on-year, accounting for more than half, exceeding chip sales revenue for the first time. Shifting from selling chips to selling algorithm licensing and services usually means the solution is deeply bound to automakers, and the stickiness of the business is much higher than simple supply. In four years, domestic intelligent driving chips have grown from an optional backup to an indispensable mainstay for automakers.
Price reduction and volume scaling are mutually reinforcing: for every step up in assembly volume, the marginal cost of sensors is diluted a bit more; for every drop in cost, the pool of affordable car models expands. Industry institutions estimate that in this round of intelligent driving cost reduction, scaled mass production contributed about 60%, while mapless technology and algorithm iteration contributed the remaining 40%. What truly brought prices down was not the generous profit concession of any single automaker, but the entire chain improving efficiency together.
There is another layer of change that is easily overlooked: the democratization of intelligent driving is crossing the boundary of powertrain types. In the past, city NOA was almost exclusively tied to NEV (New Energy Vehicle) models. Now, fuel vehicles have also joined the battle. Chery has installed city navigation pilot into traditional powertrain models under 150,000 CNY, meaning this market is no longer just a byproduct of the rising penetration rate of NEVs, but a repricing of the entire existing market. Assembly is just the starting point; the usage rate is the ultimate goal. After BYD launched its safety guarantee commitment for intelligent parking, the user usage rate of the intelligent parking function rose from 21% to 93%. This comparison shows that only when automakers are willing to underwrite parking accidents and users no longer have to worry about no one being held responsible when things go wrong, will they truly dare to hand over control to the system.
04. Supply Chain: Segments Where Volume Scaling and Price Reductions Go Hand in Hand
The volume scaling of democratization is first reflected in the perception layer. After the debate over whether to install LiDAR and how many to install, it has directly entered a period of scaled shipments: according to statistics from Gasgoo Research Institute, the installation volume of LiDAR in domestic passenger vehicles in the first four months of 2026 exceeded 1.3 million units, approaching half of the total for the whole year of 2025.
In the first half of the year, Hesai Technology (02525.HK) shipped over 1.1 million units, of which about 840,000 were ADAS products; RoboSense (02498.HK) shipped 719,200 units, a year-on-year increase of 170%, with on-hand orders exceeding 9 million units; Seyond (02665.HK) shipped about 450,000 units, a year-on-year increase of over 3.6 times, with half-year sales exceeding the whole of last year.
Along with volume scaling, prices are dropping rapidly. The average selling price of RoboSense's ADAS LiDAR has fallen to under 2,000 CNY, and the industry generally expects it to reach the 1,000 CNY level within 2026. The three manufacturers share a commonality: their sales growth rates are all far higher than their revenue growth rates, indicating that LiDAR is transforming from a high-margin technological optional accessory into a standardized component that relies on scale.
In-vehicle vision is also upgrading simultaneously. The requirements of city NOA for perception redundancy have led to a continuous increase in the number and pixels of cameras per vehicle. 8-megapixel front-view cameras are gradually becoming standard in mid-to-high-end solutions, and in-vehicle image sensors are scaling accordingly. Domestic manufacturers such as Will Semiconductor (603501) and SmartSens (688213) are regarded as the main beneficiaries in this chain. Whether it is LiDAR or cameras, the common characteristic of the sensor segment is that the growth rate of shipment volume is significantly faster than the growth rate of revenue. The popularization of hardware is exchanging unit price for scale.
A-share companies in the computing and domain control segments are the most direct beneficiaries of this round of democratization. Specifically, Desay SV (002920) achieved a revenue of 14.901 billion CNY and a net profit attributable to parents of 1.245 billion CNY in the first half of the year. Its intelligent driving domain control products supplied to automakers such as Li Auto, Great Wall, XPeng, Geely, and Xiaomi have all achieved scaled mass production. The annualized sales of newly acquired orders exceeded 22 billion CNY, a year-on-year increase of over 20%; among them, the intelligent driving business revenue was 3.918 billion CNY, a year-on-year decrease of 5.53%. Volume scaling without synchronous revenue growth is precisely a reflection of OEMs transmitting cost reduction pressure upstream, and also a part of the cost curve for the democratization of intelligent driving. Jingwei Hirain (688326) achieved an electronics product business revenue of 2.358 billion CNY in the first half of the year, accounting for nearly 90%; Huayang Group (002906) achieved a revenue of 6.923 billion CNY in the first half of the year, with the automotive electronics business accounting for over 73%.
As L3-level regulations put redundancy requirements on the agenda, the execution layer such as brake-by-wire and steer-by-wire has begun to undertake deterministic demand. Companies such as Bethel (603596) and Baolong Automotive (603197) are regarded by institutions as beneficiary segments in the evolution from L2 to L3.
05. Demand: The Last Mile from "Equipped" to "Dare to Use"
The demand-side curve is equally steep. In the first 11 months of 2025, the cumulative sales of passenger vehicles equipped with city NOA reached 3.129 million units, accounting for 15.1% of the insured volume (new car registration caliber). Roughly calculated, about one in every six or seven newly registered cars is equipped with city NOA; among them, domestic brands account for over 80%, and models with a starting price below 300,000 CNY account for nearly 70%. Volume scaling has long been occurring not only in the high-end market. Entering 2026, the downward market penetration has significantly accelerated: the assembly rate of city NOA rose from 21.8% in April to about 25% in May (Zuos Data). CICC estimates that the penetration rate of advanced intelligent driving in the 100,000 to 150,000 CNY and 150,000 to 200,000 CNY price bands has increased to about 28% and 30% respectively. The 100,000 to 200,000 CNY range has become the main battlefield for downward penetration.
But selling and actually using are two different things. Jin Yuzhi, CEO of Huawei Yinwang, previously stated publicly that currently, less than 10% of car owners have actually used advanced driver assistance. "Most people neither dare to use it nor know how to use it." In third-party field tests, the success rate of city NOA in making right turns at intersections is about 67%, and for unprotected left turns, it is about 54%. There is still a clear gap compared to experienced drivers, and most users remain in a state of human-machine co-driving with their hands on the steering wheel and feet hovering over the brake. Thus, a slightly contradictory phenomenon has emerged: many consumers are paying for the fact that the car has city NOA, rather than paying for the fact that the car's city NOA is easy to use. Configuration has outpaced actual usage.
This precisely shows that the first wave of dividends from the democratization of intelligent driving comes from the standardization at the car purchase decision end. Intelligent driving is now ranked among the top three factors in car purchase decisions. If automakers equip the functions, they can persuade a group of people to place orders. However, the true second wave of incremental growth will only be released after the usability of the functions is improved and users dare to let go. Between daring to use it and daring to let go lies user trust.
To make users dare to use it, relying solely on the car is not enough; roadside preparations are also advancing simultaneously. Data from the Ministry of Industry and Information Technology shows that over 20,000 autonomous driving test demonstration licenses have been issued nationwide, over 57,000 kilometers of test demonstration roads have been opened, and over 60,000 sets of intelligent equipment have been deployed in 20 pilot cities for vehicle-road-cloud integration. Advancing the three lines of vehicle, road, and cloud together, the usage scenarios of city NOA are extending from pilot road sections to daily commutes. Only when the functions in the car, the equipment on the road, and the responsibilities in the law are all in place, can city NOA truly transition from being usable to being dared to be used.
06. Outlook: Competition Shifts from "Having It" to "How Good It Is"
Looking back from September 2026, the three pieces of the puzzle for the democratization of intelligent driving are all in place: regulations have drawn the line of responsibility at automakers, eliminating concerns about scaled deployment; the cost curve has lowered the threshold to the 100,000 CNY level, opening up the largest price band; and domestic chips and supply chains have prepared sufficient production capacity to catch every round of volume scaling. Huatai Securities expects that in 2026, the penetration rates of highway NOA and city NOA will reach 43% and 24% respectively, and city NOA is expected to further increase in 2027. CITIC Securities estimates that the penetration rate of city NOA is expected to rise from about 14% in 2025 to 23% in 2026, while highway NOA will rise from about 15% to 16%. There are differences in caliber among institutional judgments, but the direction is consistent: city NOA is retracing the path that highway NOA took two years ago, transforming from a trial configuration to a standard threshold.
For the supply chain, the divergence after volume scaling has just begun. Huatai Securities noted in its mid-term strategy that the intelligent driving industry is experiencing a divergence of better-than-expected supply side and lower-than-expected profitability. Revenue growth without profit growth has become a phased characteristic. The source of excess returns will shift from the beta of rising penetration rates to the alpha of profit realization. Investors can focus on LiDAR with the most obvious trend of profit and loss improvement, intelligent driving chips under the rising threshold of computing power, and platform-type suppliers that use overseas expansion to hedge against price wars. Guohai Securities, on the other hand, judges that from 2025 to 2027, a three-stage path of standard formulation, transitional implementation, and comprehensive effectiveness will be formed, paving the way for L3-level access.
The affordability of city NOA is essentially cashing out the algorithms, data, and engineering capabilities accumulated over the past five years to the broadest consumer base all at once. When responsibilities have boundaries, prices are no longer out of reach, and the supply chain can handle volume scaling, the next leg of this democratization will no longer be an arms race of functional parameters, but a long-distance race of experience and trust. Whoever can make users truly dare to let go will get the ticket to the next cycle. For supply chain companies, the most certain logic is still that simple phrase: the penetration rate is still climbing, and neither production capacity nor experience has reached its peak.
- End -