Beijing 2026 confirms: ADAS has entered the compute era

The automotive industry confirmed a major shift: ADAS has entered the compute era. In this new automotive overview, Pierrick Boulay, Principal Analyst, Automotive Semiconductors at Yole Group, explores how intelligent driving is evolving from a sensor-driven market toward AI-powered, software-defined vehicle platforms.

Leveraging Yole Group’s recognized expertise in automotive semiconductors, sensing, AI computing, LiDAR, radar, and imaging technologies, this analysis is part of Yole Group’s broader automotive intelligence offering.

To gain deeper insight into market trends, competitive dynamics, and technology roadmaps, discover the flagship Automotive ADAS 2026 report and Yole Group’s dedicated automotive research collection.

At Auto China 2026, which recently closed in Beijing, the most important ADAS message was not another camera, radar, or LiDAR configuration. It was confirmation that intelligent driving is becoming a race involving compute-led platforms. Held from April 24 to May 3 under the theme “Future of Intelligence,” the show clarified where ADAS competition is moving next: away from sensor count alone and toward the ability to process, fuse, validate, and upgrade vehicle intelligence at scale.

This shift matters because the ADAS market is entering a new expansion phase. For years, growth was tied to sensor proliferation. More cameras enabled surround vision. Radar expanded from basic detection toward richer perception. LiDAR began to support higher-end, long-range functions. But as L2+, NOA, automated parking, and advanced driving functions move across more vehicle segments, the defining issue is changing. The question is no longer simply how many sensors a vehicle carries. It is whether the vehicle has the computing architecture to transform sensor data into robust, scalable driving capability.

Beijing 2026 made this transition visible. The show was less a conventional product showcase than a technology exhibition, in which the vehicle served as the carrier for AI, software, cockpit intelligence, and assisted driving. The deeper story was underneath the bodywork: in Beijing, the vehicle’s “brain” became as important as the vehicle itself.

That brain is increasingly local. Horizon Robotics used the Beijing moment to present its Starry 6P, a cockpit-and-driving integrated chip built on a 5 nm process and rated at 650 TOPS, with a memory bandwidth of 273 GB/s. A year ago, many comparisons would have been made with NVIDIA DRIVE Orin, the 254-TOPS benchmark that powered much of the previous generation of intelligent-driving platforms. That is no longer the appropriate yardstick for new programs. NVIDIA DRIVE Thor is now the more relevant reference: the platform is positioned as the follow-on to Orin, and NVIDIA’s latest developer materials specify up to 1,000 INT8 TOPS, and 273 GB/s of memory bandwidth.

Seen through that lens, Horizon’s 650 TOPS is not a low-end local alternative. It is a serious entry into the new centralized-compute class.