SAIC Motor has announced a partnership with XHEART (新芯航途) to integrate a 700 TOPS AI computing chip into its vehicles, according to PCauto. The collaboration focuses on model adaptation and vehicle-level application, with the goal of enhancing intelligent driving experiences. The companies also plan to develop next-generation chips with thousands of TOPS of computing power for future autonomous driving needs.
Beyond paper specs
The partnership will also push for upgraded energy-efficiency evaluation standards. Instead of relying solely on metrics like TOPS per watt, the companies will assess real-world performance in driving tasks, including response speed, system power consumption, and sustained operational stability. This shift aims to ensure that chip capabilities translate into tangible driving improvements.
As end-to-end AI models, vision-language-action (VLA) models, and world models evolve, intelligent driving systems must process richer information, longer sequences, and more complex reasoning. This places new demands on in-vehicle computing platforms, which must deliver timely computation while meeting constraints on power supply, cooling, cost, and long-term reliability.
SAIC and XHEART are collaborating across four areas—architecture, energy efficiency, manufacturing process, and computing power—to bring vehicle requirements into the early stages of model and chip design. A key change is the shift from adapting models to existing chips to defining chips based on model needs. This approach sets capability goals from driving tasks and translates them into computational requirements that chips and software jointly fulfill.
Advanced memory and quantization
The companies are optimizing operator fusion, parallel execution, and task scheduling to improve computational efficiency. They are also refining bus architecture and on-chip caches to ensure critical tasks receive sufficient data bandwidth. Additionally, they are preparing for advanced memory technologies like LPDDR6 and working with partners to apply 4-bit quantization to large-parameter intelligent driving models.
Larger models require higher system energy efficiency. SAIC and XHEART will jointly optimize models, chips, and vehicle cooling, allocating computing power based on road complexity to reduce unnecessary energy consumption. The upgraded evaluation standards will consider actual performance across complete driving tasks, ensuring that stronger driving capabilities perform consistently across different vehicle models.
Advanced manufacturing processes must align with production conditions. SAIC will evaluate technical solutions considering chip design, packaging, cooling, automotive-grade reliability, and manufacturing costs to ensure that gains in computing density and energy efficiency are realized in vehicle applications.
XHEART's latest chip, the X7 Ultra, offers 700 TOPS of computing power for large models. The companies will adapt and apply this platform, while also exploring chips with thousands of TOPS. They plan to develop a family of products based on a common architecture, with multiple performance tiers to suit different vehicle models and intelligent driving scales, providing computing solutions for mass production of L3/L4 vehicles.
