SDV Systems Supply Chain

Software-defined vehicle (SDV) systems are the control, compute, interface and update layers that turn an EV's separate hardware subsystems into one integrated digital platform. Software companies, semiconductor platform vendors, traditional automotive Tier-1 suppliers and Chinese OEMs all compete here for architectural control of the vehicle.

Traditional E/E (electrical and electronic) architectures used 70-100+ isolated electronic control units (ECUs), each dedicated to one function, with little cross-domain coordination and little software change after sale.

SDV architectures replace them with centralized compute domains or zonal controllers. That cuts ECU count by 60-80%, allows cross-domain software and enables over-the-air (OTA) updates that improve a vehicle throughout its life. The move from distributed ECUs to zonal to centralized compute is the main hardware supply chain change for the 2024-2030 EV generation.

E/E Architecture Transition

Each architecture generation changes Tier-1 supplier relationships, semiconductor content, software development costs and OEM control over the vehicle stack.

Architecture	ECU Count	Control Model	OEM Control	Who Benefits
Distributed ECU (Legacy)	70-150+ ECUs	One function per ECU	Low	Traditional Tier-1s
Domain Controller	20-40 ECUs + domain controllers	One controller per domain	Medium	SoC vendors
Zonal Controller	5-15 zone controllers + central compute	Zones by location + central compute	High	OEMs with in-house software
Centralized / Vehicle Computer	1-3 compute nodes + thin zone nodes	Single vehicle computer	Maximum	Software-led OEMs

Architecture Notes

Distributed ECU (Legacy): Each ECU controls one function, with minimal cross-domain coordination. The OEM integrates but does not develop most software, so traditional Tier-1s (Bosch, Continental, Denso) who own ECU software benefit.
Domain Controller: Powertrain, chassis, body, and ADAS domains each have a dedicated controller. The OEM gains domain-level software ownership, and SoC vendors (NVIDIA, Qualcomm, Mobileye) enter as domain compute suppliers.
Zonal Controller: Zones are organized by vehicle location (front, rear, left, right), with central compute for intelligence. The OEM writes the vehicle OS and cross-domain software; OEMs with in-house software capability (Tesla, BYD, XPeng, VW SDC) benefit, along with harness simplification.
Centralized / Vehicle Computer: A single vehicle computer runs all software, with a hardware abstraction layer between OS and actuators. The vehicle is a software platform and hardware is commodity. Tesla is already there, with BYD, NIO, and XPeng; emerging in VW NKP (Neue Klasse), BMW NKP, and Hyundai ccNC.

Core SDV Domain Pillars

Domain	Primary Role	Key Components
Central Compute / Vehicle Computer	Vehicle OS, cross-domain software, AI inference	Automotive SoC, LPDDR5 RAM, NVMe storage
Zonal Controllers	Zone I/O, actuator control, sensor interface	Mid-range MCU, bus bridge, power management
Chassis & Motion Systems	Braking, steering, suspension, torque vectoring	Brake-by-wire, steer-by-wire, VDC, air suspension
Smart Cabin & Digital Cockpit	Cockpit compute, displays, HMI, infotainment	Cockpit SoC, large displays, HMI software
ADAS & Autonomy Hardware	Perception for ADAS and autonomy	Cameras, 77 GHz radar, LiDAR, ultrasonics
Vehicle OS & Middleware	Runtime, service and hardware abstraction	AUTOSAR Adaptive, OEM OS, AAOS, QNX
OTA Update System	Fleet software delivery and rollback	OTA platform, modem, secure bootloader
Vehicle Networking	In-vehicle communication backbone	Automotive Ethernet, CAN FD, LIN, FlexRay

Domain Notes

Central Compute / Vehicle Computer: Runs the vehicle OS, cross-domain software, and AI inference for ADAS and autonomy on a high-performance automotive SoC (NVIDIA DRIVE Orin/Thor, Qualcomm Snapdragon Ride, Mobileye EyeQ) with LPDDR5 RAM and NVMe storage. The SoC is the highest-value SDV component; TSMC 5/4nm fabrication; single-source risk for advanced nodes.
Zonal Controllers: Local I/O aggregation, actuator control, and sensor interface for a vehicle zone, using a mid-range MCU (Renesas RH850, NXP S32, Infineon AURIX), a CAN/LIN/Ethernet bridge, and local power management. Replaces many legacy ECUs; harness simplification is the key economic driver; AUTOSAR Classic runs here.
Chassis & Motion Systems: Software control of braking, steering, suspension, and torque vectoring through brake-by-wire actuators, a steer-by-wire ECU, a VDC (vehicle dynamics controller), and an air suspension controller. Safety-critical ASIL-D requirements; Bosch, Continental, and ZF dominant; steer-by-wire enables next-gen interior design.
Smart Cabin & Digital Cockpit: Cockpit compute, display management, HMI, voice/infotainment, and passenger experience on a cockpit SoC (Qualcomm 8295, Samsung Exynos Auto, Texas Instruments Jacinto) with large-format displays and HMI software. Qualcomm 8295 is dominant in premium cockpits; Chinese OEMs (XPeng, NIO, Li Auto) are investing heavily in proprietary cockpit software.
ADAS & Autonomy Hardware: Perception hardware for driver assistance and autonomous driving: surround cameras (8+ on advanced platforms), radar (77 GHz), LiDAR (optional by approach), ultrasonic sensors, and a sensor fusion SoC. Camera supply: Sony, OmniVision; radar: Continental, Bosch, Aptiv; LiDAR: Luminar, Hesai, Innoviz; architecture choice drives the sensor BOM.
Vehicle OS & Middleware: Runtime environment, service abstraction, cross-domain communication, and hardware abstraction via AUTOSAR Adaptive (open standard), proprietary OEM OS (Tesla, BYD, XPeng), Android Automotive OS (Google), and QNX (BlackBerry). The vehicle OS is increasingly proprietary and a competitive moat; OEMs that control the OS control the feature roadmap and Tier-1 relationships.
OTA Update System: Software delivery, version management, rollback, and campaign management across the entire fleet, using an OTA platform (Aptiv/Wind River, Harman, HERE Technologies, Excelfore), a cellular/Wi-Fi modem, a secure bootloader, and delta compression. OTA is the physical connection between the vehicle and the AI training loop; every software update is a training data exchange.
Vehicle Networking: In-vehicle communication backbone between compute nodes, zones, and sensors: Automotive Ethernet (100BASE-T1, 1000BASE-T1, 10GBASE-T1), CAN FD, LIN, and FlexRay (legacy). Ethernet is replacing CAN as the primary backbone; 10Gbps links carry camera data; Marvell, NXP, and Broadcom are switch/PHY suppliers.

Key Suppliers by Layer

Central Compute SoC

Company	Stack / Platform	Notable Point
NVIDIA	DRIVE Orin (254 TOPS), DRIVE Thor (2,000 TOPS)	Leads ADAS and autonomy compute; Mercedes, Volvo, Li Auto, XPeng
Qualcomm	Snapdragon Ride Elite; 8295 cockpit	Strong in ADAS and cockpit; BMW, GM, Renault
Mobileye	EyeQ Ultra	Integrated perception and planning, ADAS-focused; Ford, Volkswagen, Nissan
Texas Instruments	TDA series	Mid-range ADAS on high-volume mainstream platforms
Renesas	R-Car series	Japanese OEM platforms; Toyota, Honda
BYD / HiSilicon (Huawei)	Proprietary Chinese OEM compute	AITO and Avatr platforms

Zonal and Safety MCUs

Company	Stack / Platform	Notable Point
Renesas	RH850, R-Car	Largest automotive microcontroller (MCU) supplier globally
NXP Semiconductors	S32 family	Leads automotive networking and zonal controllers
Infineon	AURIX	ASIL-D safety MCUs for braking, steering and powertrain safety functions
STMicroelectronics	SPC5	European OEM safety applications
Texas Instruments	Hercules TMS570	Functional safety applications

Cockpit SoC

Company	Stack / Platform	Notable Point
Qualcomm	Snapdragon 8295	Premium digital cockpit; eight screens, AI assistant, 60 TOPS
Samsung	Exynos Auto V920	BMW Neue Klasse exclusive; 50+ TOPS
MediaTek	Dimensity Auto	Entering the cockpit SoC market; strong on Chinese OEM platforms

Vehicle OS and Middleware

Company	Stack / Platform	Notable Point
BlackBerry	QNX	Safety-certified real-time operating system (RTOS) used by most Tier-1 ECU suppliers
AUTOSAR Consortium	AUTOSAR Classic and Adaptive	Open standard for ECU and adaptive platform software
Google	Android Automotive OS	Infotainment layer; Volvo, Polestar, Renault, GM
Tesla	Proprietary OS	Most mature full-stack proprietary OS; continuously updated via OTA
Volkswagen / CARIAD	VW.OS	Group SDV platform; delayed but advancing
XPeng	XGOS	Proprietary Chinese OEM OS integrated with the XNGP autonomy stack

OTA Platforms

Company	Stack / Platform	Notable Point
Harman (Samsung)	OTA middleware and cloud platform	Broad OEM adoption
Aptiv / Wind River	OTA platform	OTA integrated with vehicle networking
Excelfore	Lightweight OTA protocol	Built for constrained ECUs
Tesla, BYD, NIO	OEM-proprietary OTA pipelines	Fully in-house

SemiconductorX estimates 2,500-4,000 discrete semiconductor devices per Tesla across nine chip categories. See: Semiconductors in a Tesla (SemiconductorX)

OTA Loop, Training Clusters and Data Centers

Every OTA-capable vehicle is both a data source and a software delivery endpoint. Three links connect the fleet to AI compute:

OTA loop: the fleet generates terabytes of sensor data (camera, radar, GPS, CAN logs) per vehicle per day, and OTA updates push improved software and autonomy models back to it.
Training clusters: fleet data is ingested, labeled or used for self-supervised learning, and trains neural network models for perception, planning and control. Tesla Dojo, NVIDIA DGX clusters and Chinese OEM proprietary GPU farms do this work.
Data centers (covered on DatacentersX): host the training clusters, model serving, fleet management software and OTA delivery pipelines. AI training cluster buildout is driven directly by OEM autonomous fleet scale.

An OEM's SDV capability therefore tracks its AI training infrastructure investment. Tesla's position in FSD (Full Self-Driving) rests on its car software plus 4+ billion miles of labeled fleet data and the Dojo supercomputer trained on it. XPeng and Huawei Qiankun are building the same loop on Chinese road data.

At an estimated 2,500-4,000 semiconductor devices per Tesla, the 1.65 million vehicles Tesla produced in 2025 represent demand for 4.1-6.6 billion devices from one OEM in one year. Across a global EV market of approximately 21 million vehicles in 2025, EV semiconductor demand is estimated at 21-84 billion devices a year, before humanoid robots, autonomous equipment and drones that draw on the same supply chains.

See: AI Training Infrastructure for Autonomy | ADAS & AV Technology Stack

Chinese and Western SDV Platforms

SDV competition splits East and West, and the split shapes supply chain dependencies and geopolitical risk.

Company	Stack / Platform	Notable Point
Tesla	Proprietary OS, OTA, autonomy; HW4/HW5; Dojo	Western full-stack benchmark
BYD	Proprietary OS; DiPilot ADAS	Full-stack proprietary
XPeng	XNGP / VLA 2.0; XGOS	Most advanced Chinese autonomy stack
Huawei	Qiankun ADS	Dual-use geopolitical concern
Volkswagen	CARIAD / VW.OS; Rivian partnership	€2B+ CARIAD write-down
Mercedes, BMW	MB.OS, BMW OS 9; NVIDIA DRIVE	Behind Chinese peers on OTA cadence
GM, Ford	GM Ultifi; Ford FNV4 zonal	App-based feature unlocks

Platform Notes

Tesla: Proprietary OS, OTA, and autonomy stack; HW4/HW5 in-house SoC; Dojo training. The most mature Western full-stack SDV and the benchmark others measure against.
BYD: Proprietary OS, BMS software, thermal management, and autonomy; DiPilot ADAS platform. Full-stack proprietary with a growing software team.
XPeng: XNGP / VLA 2.0 large driving model and XGOS vehicle OS. The most advanced Chinese autonomy software stack; heavy AI infrastructure investment; exporting to Europe.
Huawei: Qiankun ADS, supplied to AITO (Seres), Avatr (Changan), and others. One of the five AV architecture approaches; a dual-use geopolitical concern for Western OEMs.
Volkswagen: CARIAD / VW.OS, with a Rivian partnership for the SDV platform. The most publicized Western SDV struggle, with a €2B+ write-down on CARIAD; Neue Klasse (2025+) represents a reset.
Mercedes, BMW: MB.OS and BMW OS 9, both adopting NVIDIA DRIVE for ADAS compute. German OEMs are behind Chinese peers on OTA cadence.
GM, Ford: GM Ultifi and Ford FNV4 zonal architecture are the US OEM SDV platforms; Ultifi aims for app-based feature unlocks.

Supply Chain Risk & Disruption

Advanced-node SoC concentration: NVIDIA DRIVE Thor, Qualcomm Ride Elite and Mobileye EyeQ Ultra are all made at TSMC on 5nm/4nm, a single-source risk with Taiwan geopolitical exposure.
Recurring automotive MCU shortages: the 2020-2022 automotive chip shortage exposed the fragility of just-in-time MCU procurement. Renesas, NXP and Infineon lead times all spike in upcycles. See: EV Semiconductor Dependencies
Software talent: OEMs need thousands of software engineers to build proprietary OS and autonomy stacks, competing with hyperscalers for the same people.
AUTOSAR fragmentation: AUTOSAR Classic and Adaptive together add integration complexity, and Chinese OEMs increasingly bypass AUTOSAR with proprietary middleware.
OTA cybersecurity: UN R155/R156 require cybersecurity management systems and OTA update certification for all new vehicles in Europe from 2024, which affects OTA platform vendors.
Chinese OEM software export controls: Huawei Qiankun ADS and HiSilicon automotive SoCs face potential export restrictions in Western markets, and OEMs are designing around that exposure.

Outlook 2026-2030

Zonal architecture as standard: zonal controllers are expected to be the production standard across mainstream platforms by 2028, with distributed ECUs designed out of new platforms.
NVIDIA Thor: DRIVE Thor at 2,000 TOPS (against 254 TOPS for Orin) puts full autonomy compute on a single chip and is expected to be the main central compute platform for 2026-2028 launches.
In-house SoCs: Tesla HW5, BYD's in-house chips and possibly others will reduce dependence on NVIDIA and Qualcomm as OEMs integrate further into silicon.
Vehicle OS competition: the contest between proprietary OEM operating systems (Tesla, BYD, XPeng) and Android Automotive will set Tier-1 software relationships for the next decade.
Fleet data advantage: OEMs with the largest OTA-connected fleets collect the most training data, a compounding autonomy advantage that is hard for new entrants to match.

The semiconductor ecosystem serving SDV compute, power electronics, sensors and networking also serves humanoid robots, autonomous equipment and drones. A disruption in SiC wafer supply, advanced-node foundry capacity or automotive MCU production hits all of them at once. See: Semiconductors in a Tesla - Cross-Sector Semiconductor Dependency (SemiconductorX)