Xpeng Motors is set to begin mass production of its in-house AI autonomous driving chip, “Turing,” in Q2 2025. Designed to reduce dependence on Nvidia, the Turing chip rivals the performance of Nvidia’s Thor (~700 TOPS), offering significant compute power and efficiency tailored for large AI models in advanced driver-assistance systems (ADAS).

Built with two proprietary neural processing units, the chip can process models with up to 30 billion parameters and boasts 20% higher utilisation than general-purpose chips. Features include dual ISPs and a safety island for improved perception and safety in challenging conditions.

The move comes amid delays and limitations in Nvidia’s Thor rollout, affecting multiple automakers. By advancing Turing, Xpeng gains control over its tech roadmap and supply chain, aiming to launch L3 autonomous driving later this year and explore L4 in 2026.

The chip supports Xpeng’s broader AI strategy, including a massive 72M-parameter “World Foundational Model,” trained in the cloud and distilled for deployment on vehicles. Xpeng also plans to use Turing in future AI robots and flying cars.