Rising Compute and Data Costs Push ADAS Toward Modular Agentic AI Architecture

Escalating semiconductor costs and compute demands are driving automakers to reconsider the implementation roadmap for advanced autonomous driving. During a panel at AutoSens Europe 2026, Duong-Van Nguyen, Global Deputy CEO of VinFast, and Igal Raichelgauz, founder and CEO of Autobrains, explained that monolithic end-to-end deep learning models expand dramatically in size and compute requirements when addressing rare corner cases, exceeding the hardware budget of most vehicle OEMs. To overcome these computational limitations, Autobrains advocated for an ADAS approach based on Agentic AI. Rather than running a single massive neural network continuously, this architecture breaks the driving task into smaller, specialized software agents that activate only when required. Raichelgauz noted that this method substantially lowers computational load and reduces required training data by approximately two to three times, cutting development costs. Nguyen added that data infrastructure presents major financial hurdles, with setup costs reaching approximately $700 million and subsequent annual maintenance requiring substantial ongoing expenditure. Furthermore, Level 3 redundancy requirements create millions of potential chip-arbitration test cases that take years to validate in real and virtual testing. Given these financial and validation constraints, automakers expect commercial rollouts over the next two to three years to center incrementally on supervised Level 2+ and Level 2++ systems.

Key takeaways

• Monolithic end-to-end models driving corner-case handling push computational demands and chip hardware costs beyond viable consumer pricing.
• Modular Agentic AI divides driving tasks into specialized agents activated on demand, reducing training data needs by 2 to 3 times and curbing compute load.
• Data infrastructure setup requires hundreds of millions of dollars, while multi-chip Level 3 validation entails millions of complex test cases.
• Due to cost and regulatory constraints, automakers anticipate that Level 2+ and Level 2++ systems will dominate commercial deployment over the next 2-3 years.