L2 시대에는 국제적인 ADAS Tier 1 공급업체들이 선점한 기술력으로 시장을 독점하고 있었습니다. 업계가 L2+ 이상 영역(L2+, L2.5, L2.9)에 진입하면서 국내 ADAS Tier 1 공급업체는 미드 레인지 컴퓨팅 칩의 성능 최적화, 엔드 투 엔드 아키텍처의 알고리즘 하드웨어 조정, 국지적 시나리오에 대한 신속한 대응이라는 세 가지 핵심 역량에 의존하고 있습니다. 핵심 역량에 의존하여 돌파구를 마련하고 점차 시장을 독점하고 있습니다.
ResearchInChina는 2024-2025년 국내 ADAS Tier 1 공급업체 25개사의 전략적 레이아웃(세계 동향, 국경을 초월한 기술), 기술 경로(칩 선택, 엔드투엔드 지능형 주행의 반복), 제품 포트폴리오 전략(지능형 주행 솔루션, 주행 및 주차 통합, 조종석 주행, 주행 및 주차 통합, 중앙 컴퓨팅 플랫폼, 기타 도메인 제어 제품)을 체계적으로 분해하여 7가지 주요 발전 추세를 정리했습니다. 7가지 주요 진화 동향을 정리했습니다.
동향 1: L2.5/L2.9, 신차 탑재율 1위로 급성장하는 시장으로 부상
증분시장(신규 출시 차량) 측면에서 2023-2025년 국내 승용차 지능형 주행 탑재 패턴은 큰 차별화 추세를 보이고 있습니다. L2.5/L2.9 지능형 주행이 비약적으로 성장하고 있습니다. 2023년 출시된 신차 중 L2.5와 L2.9 스마트 드라이빙을 탑재한 차량은 각각 4.57%와 3.3%에 불과했습니다. 그러나 2025년 1-4월에는 L2.5 스마트 드라이빙이 탑재된 신차의 비율이 30.20%, L2.9는 34.82%로 급증하며 시장 침투가 두드러집니다.
기존 L1-L2 지능형 주행 기능의 탑재율은 지속적으로 감소하고 있으며, 이는 보다 진보된 지능형 주행 기능에 대한 소비자 수요가 시장 구조를 재편하고 있음을 반영합니다.
동향2: 2024-2025년, 지능형 운전은 '고급 옵션'에서 '국가 표준'으로 보급의 '전환점'을 맞이할 것입니다.
2023년, L2.9의 ADAS 기능은 중상급 모델을 타겟으로 했습니다. 탑재율로 보면 25-30만 위안, 30만 위안 이상 모델의 탑재율이 높습니다. 특히 25만-30만 위안 모델의 L2.9 탑재율이 가장 높아 2023년 23.4%에서 2025년 1-4월에는 53.5%로 상승했습니다. 50만 위안 이상 고가 모델의 L2.9 탑재율은 2023년 7.5%에서 2025년 1-4월 32.2%로 가장 빠르게 성장했습니다.
지능형 주행 기술의 로드맵이 점차 명확해지면서 많은 OEM들이 기술 경로, 도시 커버리지, 용도, 비용 측면에서 도시 NOA에 주목하고 있습니다. 경쟁은 치열한 단계에 접어들었습니다. 2024년 이후 L2.9가 보급되기 시작했습니다. 2024년, L2.9는 우선 15만 위안에서 20만 위안 가격대의 모델로 내려가고, 10만 위안에서 15만 위안 가격대의 모델을 커버하게 될 것입니다. 이는 높은 수준의 자율주행에 대한 사용자들의 인식과 수용이 점차 높아지고 있으며, 높은 수준의 자율주행은 많은 OEM의 핵심 경쟁력도 반영하고 있습니다.
동향 3: ADAS Tier 1 공급업체들이 점차 5대 경쟁 진영을 형성하고 있습니다.
여러 차례 시장 경쟁을 거쳐 국내 스마트 드라이빙 Tier 1 공급업체들은 점차 5대 경쟁 진영을 형성하고 있습니다. 각 진영은 차별화된 기술 경로와 자원 축적에 의존하여 독자적인 시장 경쟁력을 구축하고 있습니다.
동향 4: 세계 진출 가속화, 국내 ADAS Tier 1 공급업체들의 해외 진출
동향 5: 새로운 중 연산 능력 지능형 구동 칩이 성능과 비용의 균형을 맞추는 핵심 캐리어가 될 것입니다.
동향 6: 엔드-투-엔드 모델의 진화가 가속화되면서 국내 지능형 주행 산업 체인은 '연산 능력 경쟁'에서 '알고리즘-하드웨어 협력 최적화'로 변화할 것입니다.
동향 7: Tier 1 ADAS 공급업체들은 EAI의 레이아웃을 가속화하고, 기술 동질성을 촉진요인으로 하는 두 번째 성장 곡선을 추구합니다.
중국의 ADAS와 자율주행 Tier 1공급업체에 대해 조사분석했으며, ADAS 시스템의 탑재수와 탑재율, 제품과 솔루션의 비교 등의 정보를 제공하고 있습니다.
Research on Domestic ADAS Tier 1 Suppliers: Seven Development Trends in the Era of Assisted Driving 2.0
In the L2 era, international ADAS Tier 1 suppliers monopolized the market with their first-mover technology advantage. When the industry entered the arena of L2 + and above (L2 +, L2.5, L2.9), domestic ADAS Tier 1 suppliers relied on three core capabilities to achieve breakthroughs - performance optimization of mid-range computing chips, algorithm hardware coordination of end-to-end architecture, and rapid response of localized scenarios, gradually becoming the dominant force in the market.
This qualitative change in the industry is directly reflected in the continuous expansion of domestic ADAS Tier 1 suppliers in 2022: ResearchInChina's first report domestic ADAS Tier 1 suppliers only covered 7 companies, focusing on the implementation of basic functions; in 2023, it increased to 12 companies, incorporating algorithm-driven emerging players; in 2024, it expanded to 20 companies, covering cross-border transformation players; and in 2025, it included 25 companies.
ResearchInChina systematically deconstructs the strategic layout (global expansion, cross-border technology), technical paths (chip selection, end-to-end intelligent driving iteration), and product portfolio strategies (intelligent driving solutions, driving-parking integration, cockpit-driving, driving-parking integration, central computing platforms and other domain control products) of 25 domestic ADAS Tier 1 suppliers in 2024-2025, and summarizes seven major evolution trends:
Trend 1: L2.5/L2.9 becomes the fastest-growing market segment with the highest installation rate in new cars
From the perspective of the incremental market (newly launched vehicles), the domestic intelligent driving installation pattern of passenger cars shows a significant trend of differentiation during 2023-2025. L2.5/L2.9 intelligent driving has grown by leaps and bounds. Only 4.57% and 3.3% of new models launched in 2023 were equipped with L2.5 and L2.9 smart driving, respectively. However, by January-April 2025, the proportion of new cars equipped with L2.5 smart driving had soared to 30.20%, and L2.9 34.82%, showing strong market penetration.
The installation rate of traditional L1-L2 intelligent driving functions continues to decline, reflecting that consumers' demand for higher-level intelligent driving functions is reshaping the market structure.
Trend 2: In 2024-2025, intelligent driving entered the "turning point" in terms of popularization, from a "high-end optional" to a "nationwide standard"
In 2023, L2.9 ADAS functions targeted mid-range and high-end models. From the perspective of installation rate, models with a price of RMB250,000-300,000 and above RMB300,000 had a higher installation rate. Specifically, models with a value of RMB 250,000 to RMB 300,000 had the highest L2.9 installation rate which increases from 23.4% in 2023 to 53.5% in January-April 2025. The L2.9 installation rate of high-end models worth RMB 500,000 or more grew the fastest from 7.5% in 2023 to 32.2% in January-April 2025.
As the roadmap for intelligent driving technology gradually becomes clearer, many OEMs are focusing on urban NOA in terms of technology paths, city coverage, application and costs. Competition has entered a fierce stage. Since 2024, L2.9 has begun to spread. It first descended to models priced between RMB150,000 and RMB200,000 in 2024 and further covered models priced at RMB100,000~150,000 This shows that users' recognition and acceptance of high-level autonomous driving are gradually increasing,and high-level autonomous driving also reflects the core competitiveness of many OEMs.
Trend 3: Tier 1 suppliers of ADAS gradually form five major competitive camps
After multiple rounds of market competition, domestic Tier 1 suppliers of smart driving have gradually formed five major competitive camps. Each camp relies on differentiated technology paths and resource endowments to build unique market competitiveness:
Camp 1: Local veteran Tier 1 suppliers who transfer to full-stack solutions
With Desay SV and Jingwei Hirain as typical representatives, they focus on building full-stack capabilities of "hardware-software-services" by virtue of their long-term experience in hardware research and development and stable customer base. By strengthening the development of intelligent driving software (such as perception algorithm iteration, planning and control strategy optimization), they are transforming from traditional hardware suppliers to providers of integrated hardware and software solutions, and take "full product line coverage + customer ecosystem precipitation" as a barrier to deeply bind the intelligent driving system upgrade needs of independent and joint venture OEMs.
Camp 2: International Tier 1 suppliers who propose localization
With Bosch, Continental, and Aptiv as the core, they have long dominated the domestic L2 ADAS market (such as ACC and AEB) thanks to their global technology research and development systems. As the demand for high-level assisted driving rises, they cooperate with domestic OEMs and ADAS start-ups through "technology licensing + local collaboration" to accelerate the localization adaptation of intelligent driving solutions, and simultaneously deploy high-level assisted driving scenarios such as L2+/L2.5/L2.9 to dominate the market.
Camp 3: Algorithm-driven Tier 1 suppliers
Algorithm-driven Tier 1 suppliers represented by Freetech, iMotion, Zhuoyu, QCraft, Momenta, and SenseAuto start from the core capabilities of algorithms/software (such as end-to-end planning and control, visual perception algorithms), vertically integrate sensors (such as customized cameras, lidar adaptation), domain controller hardware research and development through the "downward technology extension" strategy, and horizontally expand full-stack intelligent driving solutions. With "software-defined intelligent driving" as the core, they focus on the L2+ and higher-level markets, and gradually become mainstream suppliers by virtue of fast algorithm iteration and scenario adaptability.
Camp 4: Technology companies
Technology companies represented by Baidu and Huawei have become novel top Tier 1 suppliers in China, with full-stack software and hardware solutions.
Camp 5: Companies which conduct cross-border transformation
They entered the market from auto parts and other fields, typical examples being Huaqin, TZTEK, Keboda and other companies. In the wave of electrification, connectivity, intelligence and sharing from 2019 to 2022, they laid out intelligent driving domain controller product lines on the basis of their existing supply chain resources. Through "hardware research and development + ecological cooperation" (teaming up with algorithm companies to supplement software capabilities), they quickly entered the intelligent driving hardware market and filled the gap between traditional components and intelligent driving systems.
Each camp forms differentiated competition around technical barriers (full stack/algorithm/hardware), market positioning (L2-L2.9), and ecological synergy (OEM binding/cross-border cooperation), and jointly promotes the domestic intelligent driving Tier 1 suppliers ecology from "hardware dependence" to "technical independence" and from "functional coverage" to "deep scenario cultivation".
Trend 4: Global layout accelerates, and domestic Tier 1 ADAS suppliers embark go overseas
In the era of Assisted Driving 2.0, the speed of technology iteration and localized response capabilities have become new competitive factors. Chinese companies are breaking the traditional supply chain pattern with rapid innovation in the fields of smart cockpits, domain controllers, etc. (such as cockpit-driving integration solutions, multi-sensor fusion algorithms), as well as the cost advantages brought by large-scale production.
For domestic leading Tier 1 ADAS suppliers, relying on stable growth in cash flow and deep technical accumulation, accelerating global layout has become their core strategy to break through the market ceiling and enhance international competitiveness.
For example, Desay SV is building a globally competitive intelligent driving solution system through a combination strategy of "product adaptation + local development + ecological integration". From 2020 to 2024, its overseas revenue increased year by year with a compound growth rate of 28.9%, reaching RMB1.708 billion in 2024.
In terms of product layout in overseas markets, Desay SV and other companies have become important partners for overseas OEMs seeking "intelligent upgrades". In April 2025, it demonstrated the "chip platform + intelligent solution" collaborative model built in deep cooperation with NVIDIA and Qualcomm.
Trend 5: The emerging medium-computing-power intelligent driving chips have become the core carrier for balancing performance and cost
ResearchInChina 's survey on the new solutions of 25 domestic Tier 1 ADAS suppliers in 2024-2025 shows that L2.5/L2.9 mainly requires the computing power of 80-150TOPS (automotive-grade AI computing power), which is exactly offered by mid-range chips - compared with high-end chips (300TOPS and above), their unit computing power cost is reduced by more than 60%, and they can cover more than 90% of scenarios through algorithm optimization, accurately matching the mainstream market's demand for "uncompromising performance, affordable cost".
The current domestic intelligent driving chip market has seen a core triangle ecology consisting of Horizon Robotics, Qualcomm, and NVIDIA, but the positioning of the three is significantly differentiated:
NVIDIA Orin still dominates the high-end market (vehicles priced above RMB300,000), supporting L2+ to L4 functions in all scenarios with its CUDA ecosystem and high computing power (254-2000TOPS), but its BOM cost is high and it is difficult to sink into the mainstream market;
Qualcomm Snapdragon Ride Flex relies on the heterogeneous computing advantages of mobile chips to occupy a certain share in "cockpit-driving integration", and its automotive-grade mass production cycle and ecological adaptability should be improved;
Horizon Journey 6 (J6E/J6M) has become the core carrier of domestic substitution: more than half of the new intelligent driving solutions launched in 2024-2025 choose Horizon J6E/J6M, forming a technical path of "algorithm optimization drives computing power efficiency improvement, and hardware cost adapts to mid-range market demand", presenting a differentiated development logic of "intelligent driving algorithm addition and computing power configuration subtraction", making up for the marginal cost of computing power through algorithm innovation, and achieving a balance between performance and cost.
Trend 6: Amid the accelerated evolution of end-to-end models, the domestic intelligent driving industry chain is transforming from "computing power competition" to "algorithm-hardware collaborative optimization"
Starting from 2023, the evolution of end-to-end models has accelerated, and the domestic intelligent driving industry chain is transforming from "computing power competition" to "algorithm-hardware collaborative optimization". Major intelligent driving chip companies have promoted intelligent driving technology to a wider market through ecological layout. From the perspective of mass production and delivery, the leading Tier 1 ADAS suppliers made breakthroughs in mass production based on end-to-end model architectures and covered many models from AITO, IM, WEY Blue Mountain, Li Auto and so on in 2024-2025.
In the dimension of technological evolution, Tier 1 ADAS suppliers accelerate breakthroughs through architecture iteration and model upgrades. Huawei built an end-to-end perception base with GOD 2.0 + RCR 2.0 in 2023, and it advanced to WEWA to realize a "vehicle-cloud" behavior model closed loop in 2025. SenseAuto transitioned from "perception integration" to UniAD in 2023-2024, and launched a generative intelligent driving solution in 2025, introducing reinforcement learning. With the help of "world model + reinforcement learning", vehicles will have long-term thinking chain reasoning capabilities and make human-like decision-making. In April 2025, QCraft officially announced a one-model end-to-end solution based on a single Horizon J6M chip.
Huawei WEWA: It consists of the world engine on the cloud and the world action model on the vehicle. The world engine on the cloud side uses diffusion generation model technology to generate various extreme driving scenarios, such as bizarre accidents (for example, when there is a vehicle or obstacle blocking the view ahead, a non-motorized vehicle or pedestrian suddenly jumps out from the roadside, and the driver fails to avoid it in time, often causing an accident) and sudden braking of the front car, etc., to solve the long-tail problems that are common in the industry in the way of "AI training AI". The world action model on the vehicle is trained based on sensor data, builds a native base model, and introduces the MOE system.
SenseAuto R-UniAD innovatively introduces a world model and reinforcement learning to build a new paradigm of generative intelligent driving. The "SenseAuto Enlightenment" world model accurately simulates the physical laws, traffic rules and motion laws of the real world through 3D reconstruction and physical modeling, and builds a "virtual environment" similar to the Go board. Reinforcement learning repeatedly simulates problem scenarios, optimizes the end-to-end model, solves specific scenario problems and achieves generalization, breaks through the upper limit of human driving behavior and reduces the cost and risk of data collection. It is expected that by the end of 2025, this intelligent driving solution will be mass-produced for vehicles.
QCraft's one-model end-to-end solution based on the single journey chip: In terms of software algorithms, in order to address the safety risks caused by the black box characteristics of end-to-end technology, QCraft integrates the "space-time joint planning" experience that has been fully verified in mass production into the one model end-to-end design. In the online model training stage, a safety alignment mechanism is introduced; at the offline training level, a world model based on motion simulation is adopted. From model architecture design, training strategy, to technical infrastructure, it is ensured that the entire end-to-end model can fully play the advantages of flexibility and high upper limit, and high lower limit and higher safety can also be achieved.
Momenta's R6 flywheel model: In 2024, the one-model end-to-end solution was implemented on a large scale, and was available nationwide. In the second half of 2025, the R6 flywheel model based on reinforcement learning will be released, and reinforcement learning will be introduced for model training.
Neusoft Reach launched the next-generation two-model end-to-end solution at the end of 2024: AI Co-Driver (world restoration model) + AI Planner (decision-making & planning model)
Neusoft Reach's full-stack self-developed end-to-end model architecture is based on the AI Co-Driver. Through heterogeneous sensor fusion, the real physical world is mapped into high-dimensional vector data to provide a basis for decision-making; on the other hand, based on the AI Planner, it combines physical laws with human common sense modeling to achieve a closed loop in complex scenarios. In order to solve problems like the "black box principle", high debugging difficulty and poor explainability of AI foundation model technology, the company has established an intrinsic safety layer for foundation model applications, aiming to empower all L2 assisted driving products with functional safety mechanisms, SOTIF mechanisms and a full-process quality system for safe development that meet the safety requirements of advanced assisted driving, and significantly improve the overall safety level.
Trend 7: Tier 1 ADAS suppliers accelerate the layout in EAI and seek a second growth curve driven by technology homology
In essence, Tier 1 ADAS suppliers enter the EAI arena to build a new growth engine through "automotive intelligent technology spillover". Intelligent driving and EAI are highly reused at the level of sensors (lidar/cameras), computing power platforms (automotive-grade chips), and algorithms (path planning/motion control). Tier 1 ADAS suppliers can embark on the pan-robotics field with "automotive-grade technology". From the perspective of the products deployed, there are "hardware migration" and "software empowerment"; from the perspective of access, acquisition, investment, establishment of subsidiaries, etc. are the ways.
Analysis on iMotion's entry into EAI
iMotion has acquired related companies and anchored specific application scenarios such as charging to accurately enter the robot arena. iMotion Robot, a wholly-owned subsidiary of iMotion, went into operation in March 2025, integrating autonomous driving and robotics technology, targeting the EAI market. On May 12, 2025, iMotion Robot acquired 2/3 of Suzhou Little Craftsman Robot's shares and became the controlling shareholder of the latter.
Lenovo Vehicle Computing's layout in the pan-robotics field
Lenovo Vehicle Computing launched the pan-robotics intelligent computing platform - RH1 in May 2025. The domain controller has a DL computing power of up to 2000TOPS, supports FP4 to provide Drive-LLM support for LLMs and VLMs, accelerates edge deployment and performance optimization, and provides a solid foundation for multimodal intelligent applications. In addition, Lenovo Vehicle Computing has strategically invested in a number of pan-robotics companies through "financial support + technology output + joint development", and achieved large-scale applications in multiple fields of pan-robotics.
China's intelligent driving industry is undergoing a critical transition from technology verification to value realization. Cost optimization driven by mid-range computing chips, algorithm hardware collaboration led by end-to-end architectures, ecological breakthroughs accelerated by global layout, and technology spillover effects initiated by EAI jointly forge the core momentum of industrial upgrading.