Waymo unveils custom 5nm chip for its next-gen Robotaxis while retaining key partners

Alphabet subsidiary Waymo has published the hardware architecture of its onboard compute stack, revealing a custom 5-nanometer Application-Specific Integrated Circuit (ASIC) designed in-house to power its next-generation driverless vehicles.

The custom silicon, manufactured by TSMC, delivers over 1,000 TOPS (trillions of operations per second) of machine learning performance. While initial market reports framed the chip as a total replacement for third-party silicon, Waymo confirmed that Nvidia and AMD remain key suppliers within a heterogeneous, multi-chip compute array.

Front-End Processing for Raw Sensor Streams

Rather than running the core driving policy engine, the new ASIC acts as an accelerated front-end processor. It handles the massive influx of raw data streaming from 13 cameras, 4 lidar units, and 6 radar arrays before passing processed signals to the central perception brain.

• Temporal Denoising: Real-time cleaning and frame filtering to improve perception accuracy in low-light and adverse weather conditions.

• Sensor Fusion Acceleration: Rapid alignment of camera, lidar, and radar data streams to minimize latency in the "pixels-to-actuation" loop.

• Deployment Platform: Integrated into Waymo's 6th-generation vehicle platform, developed in partnership with Geely's Zeekr brand (the "Ojai" vehicle).

Redundancy and Power Constraints on Public Roads

Waymo emphasized that building custom silicon allows the company to meet strict automotive constraints that off-the-shelf server chips cannot address. The compute system features dual independent processing engines that run parallel workloads, ensuring full hardware redundancy if a primary circuit fails.

Additionally, by moving front-end sensor processing to a dedicated ASIC, Waymo lowers thermal output and energy consumption, preserving the battery range of its electric robotaxi fleet.

The custom chip marks a key milestone in Alphabet's broader push toward proprietary silicon, aligning Waymo's hardware strategy with Google's custom Tensor Processing Units (TPUs) to optimize operational costs as its commercial driverless service expands across major US metropolitan markets.