Volkswagen Group and Horizon Robotics are deepening their AI foundation model collaboration through CARIZON, adopting a white-box licensing model to develop a unified AI driving solution, with L3 autonomous driving targeted for delivery in China from the second half of 2027, while L2++ solutions entered mass production in Q3 2026.
Partnership Upgrade: From Procurement to White-Box Licensing
On July 22, 2026, Volkswagen Group and Horizon Robotics officially announced the deepening of their AI foundation model collaboration through joint venture CARIZON. Unlike conventional component procurement, this partnership adopts a white-box licensing model — meaning CARIZON gains access to Horizon's underlying AI foundation model capabilities, enabling engineers to customize and optimize at the底层 level rather than simply calling APIs.
Key advantages of this model:
- Technology autonomy: Volkswagen can independently iterate and optimize its AI driving system
- Deep customization space: Sufficient room for autonomous iteration of L3/L4 features
- End-to-end data-driven: Aligned with the world model evolution trend
Volkswagen Group CEO Oliver Blume stated: "China has become one of Volkswagen Group's key innovation and technology hubs, particularly in software, artificial intelligence and automated driving."
Technical Architecture: AI Model × Self-Developed Chip × Data Platform
The technical architecture consists of three core elements forming a complete intelligent driving stack:
| Layer | Core Component | Function |
|---|---|---|
| AI Algorithm | Horizon AI Foundation Model | End-to-end data-driven driving decisions |
| Chip Hardware | CARIZON C7H SoC | 500-700 TOPS, real-time sensor data processing |
| Data Platform | GAIA World Model | Simulation training and scenario generalization |
These three elements will be deeply integrated with Volkswagen's first locally developed zonal E/E architecture — CEA (China Electronic Architecture). CEA's advantages include:
- Cross-platform adaptability: Flexible deployment across multiple vehicle platforms
- Global expansion: Coverage for China and selected international markets
- Continuous upgrade capability: Common technical foundation for production ADAS upgrades
Product Roadmap: From L2++ to L3/L4
Volkswagen's intelligent driving deployment in China follows three clear phases:
Phase 1 (Completed):
- CARIZON's first ADAS delivered by end of 2025
- Mass-produced on ID. UNYX 07 sedan and ID. UNYX 06 SUV
- Features include highway navigation-assisted driving and memory parking
Phase 2 (Q3 2026):
- Full-scenario ADAS enters mass production
- Added urban navigation-assisted driving
- Rolled out across seven new electrified models from VW's three Chinese JVs
Phase 3 (H2 2027 onwards):
- L3 autonomous driving capabilities begin delivery
- System assumes driving responsibility under defined conditions
- Lays foundation for L4 driverless mobility services
Volkswagen Group China CTO Thomas Ulbrich noted: "CARIZON has always been a vital part of our local R&D efforts in China. With this new agreement, the delivery path towards L3 and L4 automated driving has become increasingly solid."
Strategic Significance: Deepening "In China, For China"
Volkswagen has invested approximately 2.4 billion euros in CARIZON, holding a 60% stake. This collaboration deepening represents a critical step in VW's "In China, For China" strategy.
From an industry perspective:
- In H1 2026, over 30% of new vehicles sold in China were equipped with advanced navigation-assisted driving
- Intelligent driving is becoming a core competitive arena
- More international automakers are turning to Chinese tech companies for support
| Automaker | Chinese Partner | Cooperation Area |
|---|---|---|
| Volkswagen | Horizon/CARIZON | Full-stack AI driving |
| Stellantis | Pony AI | Robotaxi trials in Europe |
| Mercedes-Benz | Momenta | Robotaxi services in Abu Dhabi |
For automotive industry observers in Central Asia, Russia, and global emerging markets, Volkswagen's intelligent driving technology layout in China deserves close attention. For more information, visit EX1000.COM.













