Geely-Nvidia Alliance Targets Tesla FSD

🔑 Enhanced Key Takeaways

• •The 'H7' autonomous driving scheme marks Geely's transition to a unified 'End-to-End' (E2E) neural network architecture, moving away from the previous modular perception-prediction-planning stack to better mimic human-like decision-making in complex urban environments.
• •Geely's partnership with Nvidia specifically leverages the 'DRIVE Thor' centralized car computer, which provides 2,000 TFLOPS of performance, allowing Geely to consolidate its cockpit and driving functions into a single AI brain to reduce wiring complexity and latency.
• •The 3,000-person '智驾' (Smart Drive) team integration involved merging the previously independent R&D units of Zeekr, Geely Research Institute, and Lotus Robotics to eliminate internal competition and standardize data collection protocols across all sub-brands.
• •To support the 2026 L4 target, Geely has commissioned a dedicated AI supercomputing center (Xingrui Computing Center) utilizing Nvidia DGX systems, capable of processing over 10^20 floating-point operations per second for large-model training.
• •The 43% R&D capitalization rate reflects a strategic shift to treat software as a long-term asset, a move that allowed Geely to report a 36% core profit growth despite the massive 59 billion RMB R&D expenditure in 2025.

🛠️ Technical Deep Dive

• •Compute Platform: Migration from dual-Nvidia Orin-X (508 TFLOPS) to a single Nvidia DRIVE Thor (2,000 TFLOPS) for L4 redundancy.
• •Model Architecture: Implementation of a 'World Model' for predictive path planning, utilizing Transformer-based temporal alignment to handle occlusions in urban traffic.
• •Data Loop: Automated data labeling pipeline using Nvidia Omniverse for synthetic data generation, specifically for edge-case scenarios like extreme weather and rare traffic violations.
• •Redundancy: Dual-actuator steering and braking systems integrated with the Hyperion platform to meet ISO 26262 ASIL-D safety requirements for L4 autonomy.
• •Connectivity: 5G-V2X integration for low-latency communication with smart city infrastructure, intended to supplement onboard sensors in 'blind' intersections.

🔮 Future ImplicationsAI analysis grounded in cited sources

⏳ Timeline