Gasgoo Munich- On June 17, at the "The 9th Intelligent Driving & Global Deployment Conference 2026" hosted by Gasgoo, Huawei Technologies expert Huang Ziliang noted that smart driving is fast becoming a standard feature in new energy vehicles. As penetration rates climb, the foundation of computing power is gaining critical importance.
Specifically, L2 and L2+ penetration hit 60% in 2024 and rose to 88% in 2025, with figures projected to top 95% by 2026-2027. Driving this surge are three pivotal shifts: model scales moving from millions to billions, data volumes upgrading from terabytes to petabytes, and training cycles compressing from monthly to daily iterations. Meanwhile, leading global players are ramping up computing investments, with Huawei's Qiankun ADS and Tesla's FSD shaping a dual-leadership landscape.
Furthermore, the smart driving technology roadmap has evolved through three generations: traditional multi-task models have given way to end-to-end (E2E) architectures, while VLA and world models—known for their strong comprehension capabilities—are emerging as the next core direction. Consequently, computing demands have leaped from thousands or tens of thousands of cards to the 100,000-card level, while data requirements expand from petabytes to exabytes, placing exponential growth demands on automakers' AI infrastructure.
Spanning everything from infrastructure foundations and computing platforms to L0 base models, L1 driving models, R&D toolchains, and core business scenarios, Huawei leverages its ICT base to empower the ecosystem. Ascend and the Driving SDK, backed by ultra-large clusters, full-stack AI technology, and a stable architecture, underpin the industry-leading experience of ADS.
