DeepX and Hyundai Motor Group Develop Low-Power AI Chip Platform for Robots

Gate News message, April 21 — South Korea's DeepX is partnering with Hyundai Motor Group's Robotics Lab to develop a physical AI computing platform that enables large generative AI models to run on robots in real time. The platform will leverage DeepX's DX-M2 chip, fabricated on Samsung Electronics' 2nm process, and represents the continuation of three years of joint development on low-power edge AI for robotics.

The collaboration covers low-power AI chip architecture, robot hardware, software stacks, and application libraries. Early prototypes used Nvidia's Orin AGX platform but were found to be too costly and power-intensive for widespread deployment. The partners subsequently shifted to a setup pairing Intel CPUs with DeepX neural processing units (NPUs) to reduce both cost and power consumption. The resulting edge AI chip operates at under 5 watts and has been integrated into technologies including DAL-e, a delivery robot currently in demonstration and validation phases.

The partnership reflects a broader industry shift toward specialized low-power chips for edge robotics, potentially reducing reliance on general-purpose GPUs in applications where power efficiency and thermal management are critical. Hyundai Motor Group stated the co-developed chip prepares optimized solutions for future mass-produced robots while providing greater supply chain flexibility and stability.