Waymo Reveals Its Custom Robotaxi Chip

🔑 Enhanced Key Takeaways

•The custom 5nm chip is an Application-Specific Integrated Circuit (ASIC) specifically designed for front-end processing of raw sensor data, including temporal denoising for improved perception in low light conditions.
•This ASIC delivers over 1,000 TOPS (trillions of operations per second) of machine learning performance dedicated to this initial data processing.
•Waymo's overall onboard computing system is described as a "balanced, heterogeneous system" that integrates its custom silicon with third-party CPUs, GPUs, and accelerators from companies like Nvidia and AMD for other computational workloads.
•The custom 5nm chip is manufactured by Taiwan Semiconductor Manufacturing Co. (TSMC).
•The development of this custom chip is part of a broader strategy by Waymo and its parent company Alphabet to reduce reliance on external suppliers and manage the increasing costs associated with AI computing infrastructure.

📊 Competitor Analysis▸ Show

Company	Hardware Strategy	Key Chip/Platform	Sensor Suite (Typical)	Autonomy Level (Claimed/Observed)	Notes
Waymo	Custom silicon (ASIC for front-end processing) + merchant CPUs/GPUs	Custom 5nm ASIC (1000+ TOPS for front-end), Nvidia, AMD	Lidar, Cameras (13-29), Radar (4-6)	L4 (geofenced robotaxi)	Focus on heterogeneous compute, high sensor redundancy.
Tesla	Vision-only, custom silicon	FSD HW4	Cameras (8)	L2/3 (requires supervision)	Relies heavily on neural networks and vast real-world data. Lower hardware cost (under $5,000 USD for FSD hardware).
Cruise (GM)	Custom SoC (past), Nvidia reliance (current)	Custom SoC (halted), likely Nvidia DRIVE	Lidar, Cameras, Radar	L4 (operations suspended in 2024)	Faced operational challenges and suspension of services.
Nvidia	Merchant platform supplier	DRIVE Thor (2000 TOPS)	Varies by customer	L4 (platform capability)	Provides powerful, general-purpose platforms to many AV developers.

🛠️ Technical Deep Dive

Chip Type: Application-Specific Integrated Circuit (ASIC).
Process Node: 5-nanometer.
Primary Function: Front-end processing of raw sensor data streams (lidar, radar, camera), including temporal denoising, before the data reaches the core machine learning inference engine.
Performance: Over 1,000 TOPS (trillions of operations per second) of machine learning performance dedicated to this front-end processing.
Overall Compute Architecture: A "balanced, heterogeneous system" that combines Waymo's custom silicon with third-party server-grade CPUs, GPUs, and accelerators from suppliers such as AMD and Nvidia.
Manufacturing Partner: Taiwan Semiconductor Manufacturing Co. (TSMC).
Sensor Suite (Fifth-Generation): Included a redesigned 360-degree lidar system with over 300-meter range, high-dynamic range cameras, and an imaging radar system, utilizing 29 cameras and 5 lidars.
Sensor Suite (Sixth-Generation "Ojai" Platform): Features a reduced sensor count with 13 cameras, 4 lidars, and 6 radars, leading to significantly lower hardware costs (under $20,000 per vehicle).
Lidar Capabilities: Waymo's lidar system can dynamically adjust laser pulse and beam slew rates to obtain enhanced, higher-resolution scans of specific regions or objects when needed.

🔮 Future ImplicationsAI analysis grounded in cited sources

Waymo's custom silicon strategy will accelerate its lead in Level 4 autonomous driving deployment.

In-house chip design allows for tighter hardware-software integration, optimizing performance and efficiency for specific autonomous driving tasks, leading to faster iteration and deployment.

The trend of custom silicon in autonomous vehicles will intensify, pressuring third-party chip suppliers to offer more specialized solutions.

As major AV developers like Waymo invest in custom chips for critical functions, it signals a shift towards vertical integration, potentially reducing the market for general-purpose AI accelerators in this domain.

Waymo's ability to reduce hardware costs through custom silicon will enable broader and more competitive robotaxi service expansion.

Lower per-vehicle hardware costs, as demonstrated with the 6th-generation Ojai platform, directly impact the economic viability of scaling robotaxi fleets, making services more affordable and accessible to a wider market.

⏳ Timeline

2009-01

Google Self-Driving Car Project (later Waymo) begins.

2016-12

Google Self-Driving Car Project rebrands as Waymo, an Alphabet company.

2017-02

Waymo unveils its custom-built, fully-integrated hardware suite for the Chrysler Pacifica Hybrid minivan, moving away from off-the-shelf components.

2020-03

Waymo introduces its fifth-generation Waymo Driver hardware suite, featuring redesigned lidar, cameras, and an imaging radar system, deployed on Jaguar I-Pace vehicles.

2026-02

Waymo launches its sixth-generation "Ojai" platform, significantly reducing sensor hardware costs and pushing more processing complexity into custom silicon.

2026-08

Waymo reveals its custom 5-nanometer robotaxi chip, designed for front-end sensor data processing, and identifies seven suppliers.