Report > How Software Modernization Powers Automotive Innovation

Chapter 5: Modernization

Leveraging Technology: AI and Semiconductors

Adapting to AI Demands 

Artificial intelligence is becoming a foundational automotive capability. According to our survey, 71% of respondents say AI/ML now impacts their vehicle product design, with 24% reporting extensive use and 47% citing partial integration. And while over half of organizations (55%) still use AI primarily for development tasks like code generation, simulation, and analytics, a substantial 45% now embed AI directly into their products. Most striking, 70% of teams that deploy AI in vehicles say their models are not deactivated or gated.  

This rapid shift places new demands on verification, governance, and lifecycle control. Teams must ensure that AI models—along with the datasets, calibration files, and parameters behind them—are versioned, tested, approved, and traced with the same rigor as any safety‑critical component. 

Developing Chips In-House 

The push for AI along with semiconductor shortages is leading more automotive companies to develop their own chips. For example, Rivian has introduced its Rivian Autonomy Processor (RAP1) which powers the company’s Autonomy Compute Module (ACM3). These chips, designed in-house and manufactured by TSMC, allow Rivian to: 

  • Design a fully proprietary autonomous driving system 
  • Have greater control over AI within its vehicles 
  • Match silicon closely with vehicle needs while eliminating supplier margins 
  • License its technology as a platform asset 

Tesla, BYD, and Hyundai also design chips in-house. Toyota, Stellantis, and GM have taken steps to increase their level of influence over both chip design and their semiconductor supply chain. As more companies seek more control over their products, the fundamental engineering challenge is the same: hardware, software, firmware, RTL, and configurations need to be managed as an interdependent IP portfolio with clear ownership, traceability, and reuse rules. 

How Leaders Can Meet These Challenges:

  • Keep all AI artifacts under version control (models, datasets, training scripts, logs, calibration files) so they can be reproduced, audited, and rolled back when needed. 
  • Use clear approval workflows for AI releases by tying each model to requirements, scenario tests, and acceptance criteria before deployment. 
  • Bundle models, datasets, and verification evidence together as reusable IP, so teams don’t rebuild qualification packages for every new program. 
  • Treat OTA deployment as a controlled release event, using labels, manifests, and SBOMs so updates can be traced and retracted quickly. 
  • Govern chip design IP with upstream and downstream traceability to reduce costly respins.  
  • Integrate AI and chip development into the same compliance frameworks used for functional safety and cybersecurity, ensuring audit‑ready evidence is generated automatically.