In this episode of the AI Hardware Podcast, hosts Ian Catrus and Sally Sard Fox discuss the current landscape of AI automotive chips, focusing on key players such as Intel Mobileye, Ambarella, Hailo, Nvidia, NXP, and the newcomer Athos Silicon. They begin with Intel Mobileye, highlighting its semi-independent operation from Intel, its scalable IQ chip series designed for various levels of autonomous driving, and its strategic moves to diversify beyond the Chinese market. Mobileye emphasizes a cost-effective sensor suite combining vision, lidar, and radar to enable self-driving capabilities primarily aimed at consumer vehicles and robo-taxis rather than trucks.
Ambarella is noted for its long-standing presence in vision processing, spanning consumer cameras to automotive ADAS applications. The company develops its own neural processing units (NPUs) and offers a full autonomous driving software stack, leveraging its expertise in security camera AI to enhance automotive vision systems. The hosts also discuss Hailo, a company targeting automotive AI with chips qualified for automotive standards and focusing on both vision processing and in-car user interfaces powered by small language models, enabling voice-controlled features and entertainment within vehicles.
Nvidia’s automotive chip, Thor, is presented as a significant but delayed product based on the Blackwell architecture, intended as a long-lived solution for automotive AI workloads. Despite delays attributed partly to wafer allocation favoring data center chips, Nvidia aims to grow its automotive revenue substantially. The discussion touches on the competitive landscape where OEMs often develop their own chips, leading to a fragmented market with multiple suppliers. Nvidia’s strong brand and partnerships with major automakers position it as a leading merchant supplier despite these challenges.
NXP’s IMX95 chip is highlighted as a flagship AI SoC for automotive and industrial applications, focusing on ADAS and related functionalities such as e-cockpits, connectivity, and driver monitoring. NXP has evolved from using ARM’s Ethos NPUs to developing its own AI hardware IP, reflecting its deep commitment to automotive AI. The chip targets a range of applications beyond full self-driving, emphasizing added vehicle functionalities and industrial use cases, supported by a robust software ecosystem.
Finally, Athos Silicon is introduced as a fresh entrant spun out from Mercedes, aiming to develop ASIL-D compliant chips for self-driving with advanced features like high-bandwidth memory (HBM) to meet automotive safety and redundancy requirements. The company leverages universal chiplet interconnect standards adapted for automotive use and plans a roadmap of multiple chips. The hosts underscore the challenges startups face in the automotive AI chip market due to long design cycles, stringent regulations, and shifting OEM demands, making it a tough but potentially rewarding space for innovation and investment.
