Automotive intelligent algorithms and computing power demands

In recent years, the core competition in the automotive industry has rapidly shifted from traditional engines, chassis, and manufacturing craftsmanship to 'computing power' and 'algorithms.' After the foundation of electrification was laid, intelligence became the key to success, with automotive SoCs (System on Chip) marking the starting point of this race. As advanced driver-assistance functions continue to stack up, a single SoC can no longer meet the demand. The industry is gradually moving towards higher computing power and more integrated autonomous driving platforms. On this evolutionary path, NVIDIA, Mobileye, and Huawei Ascend have each carved out distinct market approaches.

Early automotive SoCs primarily served cockpit infotainment or single ADAS functions, emphasizing low power consumption and stability. However, as the number of sensors increased and algorithms shifted from rule-based to deep learning, the demand for computing power grew exponentially. Distributed ECU architectures began to appear cumbersome and costly. Centralized or domain control architectures thus became the mainstream direction, with computing platforms required to not only process image and radar data but also support ongoing software updates. This means that chip suppliers are no longer just hardware providers; they must establish comprehensive platform capabilities that integrate both software and hardware.

NVIDIA’s approach builds on its successful experience in the data center and AI training fields. From the early Drive PX to the later Orin and Thor platforms, its core concept has been to leverage highly versatile GPU computing power to meet the needs of everything from L2 to L4 and even higher-level autonomous driving. The advantage of this strategy lies in extremely high software flexibility, allowing car manufacturers to continuously upgrade algorithms on the same platform, thereby extending the lifecycle of vehicle models. However, general-purpose computing power also means greater cost and power consumption pressures. For automakers pursuing scaled production and gross margin control, how to balance performance and cost becomes a key consideration in deciding whether to adopt NVIDIA's solution.

By comparison, Mobileye has chosen a more 'specialized' path. Its EyeQ series of chips have long focused on ADAS and perception algorithms, achieving high efficiency and reliability within limited computing power through highly customized accelerators. This gives Mobileye strong competitiveness in the L2/L2+ market and allows it to quickly form deep integration with traditional automakers. Huawei Ascend represents a different approach: emphasizing self-control, Ascend focuses on AI computing power, leveraging Huawei’s accumulated strengths in communications, systems, and software, attempting to build a complete chain from chips to full vehicle solutions. This model is particularly attractive in the Chinese market but also faces challenges in ecosystem maturity and international expansion.

The competition ranging from in-vehicle SoCs to autonomous driving computing platforms involves a comprehensive comparison of chip performance, ecosystem, cost structure, and market positioning. The future winner may not depend on a single technical metric but rather on who best understands the true needs of automakers and the market for 'smart vehicles.'