💰钛媒体•Stalecollected in 3h
EV manufacturers shift to proprietary self-driving tech
💡See how EV makers are using in-house AI to disrupt the automotive market.
⚡ 30-Second TL;DR
What Changed
Shift from third-party solutions to in-house R&D
Why It Matters
The trend suggests a competitive landscape where software differentiation is the primary driver of value. Builders should watch for the commoditization of AI-driven driving features.
What To Do Next
Evaluate the feasibility of building proprietary AI models versus licensing third-party solutions for your product stack.
Who should care:Developers & AI Engineers
Key Points
- •Shift from third-party solutions to in-house R&D
- •Democratization of premium tech across vehicle price tiers
- •Increased focus on proprietary software stacks
🧠 Deep Insight
Web-grounded analysis with 33 cited sources.
🔑 Enhanced Key Takeaways
- •The shift to in-house development by Chinese EV manufacturers is partly driven by geopolitical tensions, aiming to reduce reliance on foreign technology, particularly in high-end autonomous driving chips.
- •Vertical integration in self-driving tech allows Chinese manufacturers to achieve significant cost savings, enabling them to offer ADAS-equipped vehicles at wholesale rates 15-22% below comparable international models, bolstering their export competitiveness.
- •Advanced driver-assistance systems (ADAS) are rapidly becoming standard across various price points in China, with high-level features previously exclusive to premium models now available in vehicles priced as low as 70,000-150,000 RMB (approximately $9,800 - $21,000 USD).
- •Leading Chinese tech companies like Huawei are employing sophisticated AI architectures, including cloud-based world models, multi-agent game theory, and generative AI for scenario simulation, to accelerate the development and training efficiency of their autonomous driving systems.
- •Chinese consumers exhibit a high degree of readiness and trust in autonomous driving technology, with approximately 85% comfortable with self-driving that does not require human supervision, which contrasts with lower acceptance rates in Western markets.
📊 Competitor Analysis▸ Show
Competitor Analysis: Chinese EV Proprietary Self-Driving Systems
| Feature / Company | XPeng (XNGP) | Li Auto (AD Max / AD Pro) | NIO (NOP+) | Huawei (ADS) | BYD (God's Eye) |
|---|---|---|---|---|---|
| Key Chip(s) | Dual Nvidia Orin X (508 TOPS) | AD Max: Dual Nvidia Orin-X (508 TOPS), upgraded to Thor-U (700 TOPS); AD Pro: Horizon Journey 6M (128 TOPS) | Proprietary Shenji NX9031 (5nm, >50 billion transistors) | Self-developed processing platform (ADS 3.0: 400 TOPS) | Proprietary system (details less public) |
| Perception Hardware | 2 LiDARs, 5 mmWave radars, 11 cameras (8MP HD), 12 ultrasonic sensors | AD Max: 1x 128-line LiDAR, 6x 8MP cameras, 5x 2MP cameras, 1x mmWave radar, 12x ultrasonic sensors | Multi-sensor fusion (specifics for Shenji NX9031 not fully detailed) | ADS 3.0: 1x 192-line LiDAR, 1x 4D mmWave radar, 12x ultrasonic sensors; ADS 4.0: high-precision solid-state laser radar + distributed millimeter-wave radar | Vision-fusion (details less public) |
| Software Architecture | XNet (neural network-based perception), closed-loop self-evolving AI and data system | End-to-end + VLM, VLA driver model (AD Max) | SkyOS unification, OTA-driven improvements | ADS 3.0: GOD perception network + PDP decision network + instinctive safety network, end-to-end large model; ADS 4.0/5.0: WEWA architecture, cloud-based world model, generative AI, online reinforcement learning | "God's Eye" ADAS platform |
| Key Capabilities | Full-scenario ADAS (highways, city roads, parking), map-free navigation expansion | Full-Scenario NOA (highways, expressways, urban roads), intelligent parking, active safety (AEB, AES) | Navigate on Pilot Plus (NOP+), highway and urban capabilities | Map-free global navigation (urban/highway NCA), improved decision response speed, enhanced safety redundancy (AEB, EAS) | High-level smart driving system, standard on many models, pushing ADAS to lower price brackets |
| Pricing/Availability | Max version supports city NOA, Pro version highway NOA; available in 243 cities | AD Max/Pro available on Li L6, L9, L7, Li MEGA | NOP+ on select models; ET9 to feature Shenji chip | ADS SE deployed in 150,000 yuan segment; ADS Pro for urban NCA in 150,000 yuan models; ADS 3.0 standard on Aito Wenjie M5 Ultra | Standard on 21 models (Dynasty & Ocean series), including 100,000-150,000 RMB price bracket |
| Competitor Comparison | Direct competitor to Tesla FSD in China | Aims for better decision-making in complex scenarios and user preferences via VLA models | Investing in in-house chips to reduce Nvidia reliance and improve margins | Positions itself as an "electronic screw" empowering automakers, not a whole-vehicle manufacturer | Vertical integration provides cost advantages over foreign competitors |
🛠️ Technical Deep Dive
- XPeng XNGP: Utilizes dual NVIDIA Orin X SoCs providing 508 TOPS of computing power. Its perception suite includes 2 LiDARs, 5 mmWave radars, 11 high-definition 8-megapixel cameras, and 12 ultrasonic sensors. The system is built on a new software architecture called XNet, backed by a closed-loop, self-evolving AI and data system for continuous improvement.
- Li Auto AD Max: Features dual NVIDIA DRIVE Orin-X chips with a combined computing power of 508 TOPS, later upgraded to NVIDIA Thor-U with 700 TOPS. It incorporates a 128-line LiDAR, 11 high-definition smart driving cameras (6x 8MP, 5x 2MP), a forward millimeter-wave radar, and 12 ultrasonic sensors. The system supports end-to-end + VLM (Vision-Language-Model) and is capable of running a VLA (Vision-Language-Action) driver model.
- NIO's Proprietary Chip (Shenji NX9031): An in-house developed 5nm autonomous driving chip with over 50 billion transistors, designed to boost AI processing for safety and perception, reducing reliance on external suppliers like Nvidia.
- Huawei ADS (Qiankun ADS): ADS 3.0 uses a GOD perception network + PDP decision network + instinctive safety network, integrating a BEV solution with GOD to form a large network and adopting an end-to-end large model. It has 400 TOPS processing power and features a 192-line laser radar (250m detection range) and a 4D millimeter-wave radar. ADS 4.0 introduced the WEWA (World Engine + World Action Model) architecture, leveraging generative AI for scenario simulation. ADS 5.0 further refines this with WEWA 2.0, a cloud-based world model grounded in multi-agent game theory, enabling online reinforcement learning and applying a Safety Risk Field theory to quantify risks. It also uses a Lingqu Bus to reduce in-vehicle signal latency by 30%.
- Horizon Robotics: Offers the Horizon SuperDrive™ (HSD) full-stack solution, which is a one-stage end-to-end driving assistant. Their Journey™ 6 series processing hardware underpins these solutions. They are also developing the "Xingkong" chip, China's first cockpit-driving fusion agent chip, consolidating two domain controllers and memory systems into one to reduce costs.
🔮 Future ImplicationsAI analysis grounded in cited sources
Chinese EV manufacturers will achieve global leadership in advanced driver-assistance systems (ADAS) technology.
Vertical integration, rapid innovation cycles, strong government support, and high domestic adoption rates provide a significant competitive advantage, enabling them to outpace international rivals in deployment and feature development.
The cost of advanced self-driving features will continue to decrease, making them standard even in entry-level EVs.
In-house development and vertical integration significantly reduce per-vehicle costs, as demonstrated by companies like BYD and Huawei pushing high-level ADAS to lower price tiers.
The industry will increasingly adopt end-to-end AI models and advanced simulation environments for rapid iteration and deployment of autonomous driving systems.
Companies like Huawei and Li Auto are already implementing end-to-end large models and leveraging generative AI for scenario simulation and online reinforcement learning, indicating a trend towards more sophisticated and efficient AI-driven development.
⏳ Timeline
2017-12
NIO launched ES8, featuring L2 self-driving with Mobileye EyeQ4 chip, marking early ADAS adoption in Chinese EVs.
2022-06
Li Auto unveiled the Li L9, featuring its flagship, full-stack self-developed ADAS, Li AD Max, powered by dual NVIDIA Orin-X chips.
2022-09
XPeng released its second-generation intelligent assisted driving system, XNGP, showcasing in-house full-stack R&D capabilities.
2023-04
Huawei launched ADS 2.0, an AI-powered ADAS, emphasizing reduced reliance on high-precision maps and self-developed processing.
2025-02
BYD announced its 'God's Eye' high-level smart driving system would be standard on 21 models, democratizing advanced ADAS to lower price tiers.
2025-10
NIO confirmed its upcoming ET9 would feature the proprietary Shenji NX9031 ADAS chip, highlighting its in-house semiconductor development.
📎 Sources (33)
Factual claims are grounded in the sources below. Forward-looking analysis is AI-generated interpretation.
- digitimes.com
- mdpi.com
- zacarmate.com
- asianews.network
- eeworld.com.cn
- news18a.com
- automobility.io
- xpeng.com
- wikipedia.org
- liauto.ph
- liauto.com
- lixiang.com
- liauto.com
- portauto.com
- pestel-analysis.com
- matrixbcg.com
- counterpointresearch.com
- youtube.com
- youtube.com
- cgsi.com.sg
- bitauto.com
- eeworld.com.cn
- moomoo.com
- sailtoday.org
- reddragoncanoeclub.org
- chinadaily.com.cn
- recodechinaai.com
- qz.com
- gasgoo.com
- horizon.auto
- prnewswire.com
- berkeley.edu
- medium.com
📰
Weekly AI Recap
Read this week's curated digest of top AI events →
👉Related Updates
AI-curated news aggregator. All content rights belong to original publishers.
Original source: 钛媒体 ↗