Intelligent competition is reshaping China's automotive industry. As L2 advanced driver-assistance systems (ADAS) accelerate in popularity and city NOA (Navigate on Autopilot) expands into mainstream price segments, industry competition has moved beyond merely "having the features" to a comprehensive evaluation of mass production scale, complex scenario experience, cost efficiency, and safety responsibilities. High-level intelligent driving is shifting from a configuration selling point to a core vehicle capability, and the industry is transitioning from a technology demonstration phase to a new stage of large-scale implementation.
Based on this, this report focuses on L2 and above, conducting research from four dimensions: industry overview, market and technology analysis, trend insights, and representative manufacturers. It systematically reviews the commercialization paths of intelligent driving classification and cost structure, market penetration and competitive landscape, perception and algorithm evolution, L3/L4 pilots, industry ecosystem, and typical enterprises.
High-level intelligent driving is accelerating its downward market penetration, making large-scale capabilities a new market threshold. Data shows that from January to May 2026, the penetration rate of L2 and above ADAS features in domestic passenger vehicles has reached approximately 70%, with related features extending to vehicle models priced around 100,000 RMB. As highway NOA, city NOA, and parking features gradually become popularized, users' criteria for evaluating intelligent driving have shifted from parameters and configurations to actual availability rates, takeover frequency, OTA speed, and experience stability.
Market expansion does not mean that all players benefit simultaneously. In-house solutions by automakers hold a significant share, while the third-party market is further concentrating among leading suppliers with capabilities in mass production delivery, data closed-loop, and multi-model reuse. Future competition will shift from single-point project nominations to platform-based replication: those who can deliver stably across more brands, more price segments, and more complex roads are more likely to remain in the game.
The algorithmic system is evolving from modular rule chains to end-to-end models and large models. BEV (Bird's Eye View), Transformer, and Occupancy networks enhance spatial perception capabilities. VLA (Vision-Language-Action) further integrates vision, language, and driving actions, while world models handle scenario generation, future prediction, and closed-loop training. The core change in the technological route is to reduce the system's reliance on manual rules, gaining stronger understanding and generalization capabilities in complex intersections, interactive games, and long-tail scenarios.
The focus of perception hardware upgrades has also shifted from "stacking quantities" to "improving information quality." Combining sensor BOM costs and technology trends, the report analyzes the evolution directions of high-pixel cameras, 4D mmWave radar, and solid-state and chip-based LiDAR. As central computing and zonal control reconstruct data links, the focus of industrial competition will further shift towards the efficient collaboration of sensors, algorithms, and computing platforms. Multi-modal fusion will become the core technological route for high-level intelligent driving, gradually converging towards a central computing architecture.
L3/L4 pilots are pushing the industry into a parallel stage of "technology verification + responsibility verification." L2 safety baselines, driver monitoring, and data recording requirements are continuously being strengthened, while L3 and above autonomous driving further clarify operational design domains (ODD), minimum risk strategies, fault handling, and event tracing. Pilots in Beijing, Chongqing, and other regions, along with Robotaxi operations, indicate that the implementation of autonomous driving requires not only verifying the upper limits of functionality but also simultaneously verifying regulations, insurance, operations, and closed-loop responsibilities.
The industry landscape is converging, but what is truly being screened is systemic capability. Long-term competitiveness stems from five aspects: the training flywheel formed by real-world road data, cross-model mass production and delivery capabilities, controllable hardware and computing power costs, continuous OTA iteration efficiency, and a responsibility system covering functional safety, SOTIF (Safety of the Intended Functionality), and user services. The relationship between automakers and technology companies is also shifting from simple procurement to equity binding, joint venture co-creation, and long-term business synergy.
The report also further analyzes the overseas expansion paths of China's intelligent driving industry, covering the export of complete vehicles and intelligent driving capabilities by OEMs, the overseas expansion of solutions by technology companies, and the entry of sensor enterprises into the global supply chain. Meanwhile, taking BYD's ADAS fallback service as an example, it observes the extension of industry competition towards responsibility assumption and user protection. Through the cases of Huawei ADS 5 and Momenta, it analyzes the commercialization paths of platform-based delivery, data closed-loop, multi-model reuse, and the coordinated evolution of L2 and L4.
Overall, the ADAS industry is moving from "technically usable" to "scalable and reproducible, with assumable responsibilities." Future leaders may not necessarily possess the highest single parameters, but rather enterprises that can connect data, algorithms, complete vehicle platforms, supply chains, and safety compliance into a stable business closed-loop.