The core battlefield of intelligent driving has long shifted from “sensor stacking” to “chip computing power competition.” As the “brain” of the vehicle, intelligent driving SoC chips are responsible for core tasks such as environmental perception, decision planning, and multi-sensor data fusion, directly determining the upper limit of the autonomous driving experience. In 2025, with high-level intelligent driving penetrating into vehicles priced at 100,000 yuan and breakthroughs in domestic chip technology, the intelligent driving SoC industry is ushering in a dual resonance of “technological iteration + demand explosion,” making the wave of domestic alternatives unstoppable.
1. Intelligent Driving SoC: The “Computing Heart” of Intelligent Driving, How Important Is It?
The intelligent driving SoC (System on Chip) is a “super platform” that integrates multiple heterogeneous computing units such as CPU, GPU, NPU (or BPU), and ISP onto a single chip. Unlike MCU chips that are responsible for single control tasks, it can support multi-task concurrency and high-speed processing of massive data, making it essential hardware for L2+ and higher-level autonomous driving.
Compared to MCUs, the computing power of intelligent driving SoCs is more than an order of magnitude ahead: MCUs have a clock frequency of only 16MHz-300MHz and less than 10MB of memory, while intelligent driving SoCs can reach clock frequencies of 1GHz-3GHz+, with memory exceeding 1GB. AI computing power is measured in TOPS (trillions of operations per second), easily handling complex algorithms such as image recognition and Transformer large models.
Automotive-grade requirements are stringent: they must operate stably in extreme environments from -40℃ to 125℃ for over 15 years, with zero fault tolerance, and must also have special designs for lightning protection, anti-interference, and enhanced heat dissipation, with thresholds far exceeding consumer-grade chips.
Application scenarios are clear: low-computing power chips (2.5-20 TOPS) are suitable for front-view integrated machines, achieving L0-L2 level assisted driving; medium computing power (20-80 TOPS) supports high-speed NOA; high computing power (≥100 TOPS) is the core of urban NOA and integrated cockpit driving, used in high-end models priced over 250,000 yuan.
2. Market Space Explosion: Scale Exceeding 100 Billion by 2028, Domestic Alternatives as Core Increment
The surge in intelligent driving penetration directly ignites the market demand for intelligent driving SoCs, and the industry growth curve has entered a steep phase:
The overall market is rapidly expanding: in 2023, the domestic automotive-grade SoC market size reached 26.7 billion yuan, with a compound annual growth rate of 42% from 2019 to 2023, far exceeding global levels; it is expected to exceed 102 billion yuan by 2028, nearly tripling in five years.
High-level intelligent driving becomes the growth engine: the ADAS SoC (assisted driving) market in 2023 is 14.1 billion yuan, while the more advanced ADS SoC (autonomous driving), although still in its infancy, is expected to grow to 25.7 billion yuan by 2030, becoming the largest incremental track.
Penetration rate approaching saturation: in 2023, the penetration rate of domestic autonomous passenger vehicles reached 74.7%, and it is expected to rise to 93.5% by 2028, with almost all new cars equipped with intelligent driving functions, providing a massive demand base for SoC chips.
3. Three Major Technical Trends: Reshaping the Competitive Landscape of Intelligent Driving SoCs
The intelligent driving SoC industry is undergoing dramatic technological changes, with three major trends determining the core discourse power of future competition:
1. Integrated Cockpit and Driving: The Ultimate Fusion from “Dual Chips” to “Single Chip”
The automotive electronic architecture is evolving from distributed to centralized integration, with “integrated cockpit and driving” becoming a clear direction—integrating intelligent cockpit and intelligent driving functions onto a single SoC, achieving shared computing power and data intercommunication.
Core advantages: hardware costs reduced by 40%, significantly shortened communication latency, and support for innovative functions such as AR navigation, making it the mainstream form of the future.
Representative players: NVIDIA Drive Thor, Qualcomm SA8775P, Horizon Journey 6 series, and Black Sesame Intelligence C1200 are all targeting integrated cockpit solutions, with multiple mass-produced models expected to be launched in 2025.
2. High Computing Power + Algorithm Adaptation: Breaking Through the Bottleneck of Transformer Models
As BEV + Transformer + OCC becomes the mainstream algorithm for urban NOA, intelligent driving SoCs face new challenges of “memory access intensity”—the memory bandwidth and capacity requirements of Transformer models far exceed those of traditional CNN algorithms.
Path to technological breakthroughs: integrating Transformer engines in hardware (such as NVIDIA Hopper GPU), optimizing key operators like Layer-norm in software, and adopting three-level cache reconstruction in storage to break through the “memory wall” limitations.
The computing power race is intensifying: domestic chips have achieved leapfrog breakthroughs, with Horizon J6P reaching 560 TOPS, and Chipone Technology’s “Star One” capable of achieving 2048 TOPS through multi-chip cascading, comparable to international top levels.
3. Hardware-Software Collaboration: Open Ecosystem as Core Barrier
The traditional “hardware first, software later” serial development model is outdated, and the integrated model of “chip + software stack + development platform” has become the industry standard.
Ecological advantages determine adaptation efficiency: NVIDIA monopolizes the high-end market with its CUDA ecosystem, while Horizon has launched the “Tiangong Kaiwu” toolchain, supporting plug-and-play algorithms, significantly lowering the adaptation threshold for car manufacturers.
Parallel development shortens cycles: by implementing virtual platform licensing and software white-box solutions, hardware design and software development can be advanced simultaneously, shortening product landing cycles by 1-2 years.
4. Competitive Landscape: Overseas Dominance Gradually Retreats, Domestic “Three Tiers” Rise
The current intelligent driving SoC market shows a pattern of “overseas leaders dominating, domestic players rapidly catching up,” but domestic alternatives have entered an acceleration phase. In January-February 2025, domestic market share data shows that NVIDIA’s Orin series accounts for about 50%, while domestic manufacturers such as Huawei and Horizon have collectively exceeded 20%, with the gap rapidly narrowing.
1. Overseas Camp: Monopolizing High-End, Ecosystem is King
NVIDIA: With the high computing power of Orin-X/N chips and the CUDA ecosystem, it occupies the market for high-end models priced over 250,000 yuan, with clients including Li Auto, Xpeng, and NIO.
Mobileye: Focused on the mid-to-low-end ADAS market with EyeQ4/EyeQ5 chips, it remains a major supplier for front-view integrated machines due to its mature algorithm solutions.
Qualcomm: Entering the integrated cockpit track with SA8295/SA8775P, rapidly expanding partnerships with automakers based on its technology accumulation in the consumer electronics field.
2. Domestic Camp: Three Tiers Breakthrough, Full Bloom
First Tier (Leading Breakthrough): Horizon (Journey series chips shipped over 10 million units, with NOA chip shipments in H1 2025 expected to increase fivefold), Huawei (Ascend 610 chips equipped in models like Wenjie and Zhijie, with a market share of 9.8%).
Second Tier (Rapid Follow-Up): Black Sesame Intelligence (A1000 series deployed in Lynk & Co 08, Hozon Neta V09), Chipone Technology (“Star One” expected to be mass-produced in 2025, competing with dual Orin-X).
Third Tier (Segmented Entry): Aisin Yuanzhi (44.83% market share in the front-view integrated machine market, ranking first), Renesas (focusing on the low-computing power ADAS market).
3. Automaker Layout: Parallel Self-Development + External Procurement
Tesla insists on self-developing FSD chips, while NIO and Xpeng have also launched self-developed chips, but most automakers choose a hybrid model of “self-developed algorithms + external chip procurement” to ensure differentiation while reducing R&D risks. Traditional automakers like Geely, Changan, and SAIC are binding domestic chip manufacturers through joint ventures and strategic investments to accelerate the localization process.
5. Industry Summary: Domestic Alternatives + Technological Iteration, Opening a Golden Five Years
The intelligent driving SoC industry is standing at the intersection of “demand explosion + technological breakthroughs + domestic alternatives,” and the next five years will be a critical period for determining the industry landscape. The core trends can be summarized in three points:
Domestic alternatives are penetrating comprehensively from “mid-to-low-end” to “high-end”: the low-computing power market has achieved domestic dominance (over 50% share in front-view integrated machines), the mid-computing power market is rapidly capturing share with “cost-effectiveness + localized services,” and the high-computing power market is breaking the overseas monopoly through technological breakthroughs, with domestic share expected to exceed 50% by 2026.
The technological route is converging towards “integrated cockpit + high computing power + open ecosystem”: integrating cockpit and intelligent driving functions onto a single chip is the ultimate form, with 3nm process and Chiplet technology further enhancing computing power density, while ecological adaptation capability will become the core competitiveness of chip manufacturers.
Market competition is shifting from “computing power comparison” to “value competition”: as the issue of computing power surplus becomes apparent, automakers will focus more on the energy efficiency ratio, algorithm adaptability, and cost control of chips, rather than simply pursuing computing power values. Manufacturers with integrated capabilities of “hardware + software + services” will prevail.
6. Future Outlook: The Era of Equal Rights in Intelligent Driving, Chips as the Core Pass
As the “Moore’s Law” of intelligent driving hardware costs halves every two years, equipping 100,000 yuan models with urban NOA will become the norm, and the demand for intelligent driving SoCs will spread from high-end models to mass-market vehicles. Domestic chip manufacturers, leveraging localized services, rapid iteration capabilities, and cost advantages, are expected to achieve a “curve overtaking” in this industrial transformation.
In the future, intelligent driving SoCs will not only be the “computing heart” of vehicles but will also become the core label for differentiated competition among automakers. As technology continues to evolve, the deep integration of chips and algorithms will drive autonomous driving from “assisted” to “autonomous,” ultimately reshaping the travel ecosystem.