ADAS/AV Tech Stack Overview

Advanced Driver Assistance Systems (ADAS) and Autonomous Vehicle (AV) technologies require a multi-layered supply chain that blends semiconductors, sensors, software, and connectivity into mission-critical driving intelligence. The stack can be viewed in two dimensions: the core autonomy layers that enable safe self-driving, and the vehicle digital layers that provide connectivity, updates, and consumer experience. Both depend on fragile, globally distributed supply chains.

AI Training & Simulation

The AI video model training layer is a critical but often overlooked part of the ADAS/AV ecosystem. It is an upstream supply chain dependency and requires: 1) Petabytes of labeled driving video data; 2) Simulation for safety edge cases; 3) Enormous GPU/TPU cluster demand. Without it, inference chips in the car don’t have usable neural nets to run.

Training compute is concentrated in the US (Tesla Dojo and Nvidia clusters) and China (Baidu, Pony.ai). It is very costly and time-consuming to replicate at national scale.

Core Autonomy Layers

This is the Driving Intelligence Stack.

Perception
Cameras, LiDAR, Radar, Ultrasonic, Thermal sensors
Notes: Multi-modal sensing, object detection, redundancy.

Localization & Mapping
GNSS, IMUs, HD Maps, V2X beacons
Notes: Precise vehicle positioning, real-time updates.

Compute Hardware
AV SoCs, GPUs, NPUs, FPGAs, Edge AI chips
Notes: Nvidia Orin, Mobileye EyeQ, Tesla HW, Qualcomm Snapdragon Ride.

Domain Control Units (DCUs)
Centralized compute hubs replacing dozens of ECUs.
Notes: Bosch, Continental, Aptiv, ZF modules.

AI / ML Algorithms
Perception, prediction, planning, reinforcement learning. Custom designs per application.
Notes: Core decision-making brain for AVs.

Control & Actuation
Drive-by-wire steering, braking, throttle, powertrain.
Notes: Redundant, safety-critical actuation systems.

Safety & Cybersecurity
ISO 26262, ASIL-D systems, intrusion detection.
Notes: Redundant compute, secure communication, fail-safe modes.

Vehicle Digital Layers

This is the Connectivity & User Experience Stack.

Connectivity / Telematics
4G/5G, C-V2X, DSRC, satellite
Notes: Always-on links, fleet monitoring, V2X coordination.

OTA (Over-the-Air Updates)
Firmware, software, security patches
Notes: Tesla pioneered; now an industry standard.

ADAS / AV Platforms
Proprietary autonomy software layers
Notes: Tesla FSD, Mercedes Drive Pilot, GM Super Cruise, Ford BlueCruise, Baidu Apollo.

Middleware / OS
RTOS, AUTOSAR, ROS, AI frameworks.
Notes: Safety-critical runtime, integration of heterogeneous systems.

Infotainment Systems
IVI hardware/software, media & apps.

Notes: Android Auto, Apple CarPlay, OEM ecosystems.

Vehicle Apps & Ecosystems
Consumer-facing applications, app stores.
Notes: Navigation, streaming, smart home integration.

HMI (Human-Machine Interface)
Displays, AR HUDs, voice/gesture input.
Notes: Driver fallback (L2/L3) & passenger UX (L4/L5).

China’s 2025 Ban on Nvidia: Impact on EV Autonomy

In September 2025, China directed its automakers and AI firms to halt purchases of Nvidia chips and adopt domestic alternatives, chiefly Huawei’s MDC (Mobile Data Center) series.

Global split: U.S./EU EVs will continue with Nvidia Drive Thor, Qualcomm Ride Flex, and Mobileye EyeQ, while Chinese EVs will pivot to Huawei MDC and Ascend-based systems.
Fleet impact: This divergence will shape the robotaxi and AV truck sectors, with different inference silicon stacks powering autonomy.
Supply chain lens: The ban accelerates the bifurcation of the EV/AV autonomy ecosystem into U.S.-aligned (Nvidia/Qualcomm/Mobileye) vs. China-aligned (Huawei/SMIC) technology blocs.

Supply Chain Bottlenecks/Takeaways