Waymo, a Google Subsidiary, Mass Produces Self-Developed 1,000 TOPS Autonomous Vehicle Chip
Waymo, a subsidiary of Google, has begun mass production of its self-developed dedicated chip for the latest generation of autonomous taxis, aimed at enhancing response and navigation capabilities in complex urban environments.
The chip is an application-specific integrated circuit (ASIC) with a computing power exceeding 1,000 TOPS, which means over one trillion operations per second; based on this metric, its performance is comparable to NVIDIA's latest autonomous driving systems.
Waymo states that this ASIC handles the front-end processing of sensor data, cleaning and fusing information from cameras, LiDAR, and radar, which is then processed by the vehicle's computing platform to understand and respond to the environment, thereby accelerating the autonomous vehicle's processing speed regarding road participants and urban conditions.
The chip is manufactured using TSMC's 5-nanometer process. Waymo has not disclosed the cost per chip, production capacity, foundry order amounts, chip area, or specific power consumption parameters; while 5-nanometer is not TSMC's most advanced node, it is still widely used for high-performance chip mass production.
Waymo has not completely replaced external suppliers with its self-developed chip. Its computing architecture still includes partners such as AMD, Micron, NVIDIA, Samsung, SanDisk, Socionext, and TSMC; the self-developed ASIC is just one part of the onboard computing system, and specific component roles have not been disclosed.
In terms of market mechanisms, Waymo has integrated high-frequency, low-latency, and highly customizable perception processing into its self-developed chip, which can reduce long-term reliance on general-purpose autonomous driving computing platforms and dilute design costs as the fleet expands; TSMC benefits from the demand for custom chip manufacturing, while external chip vendors like NVIDIA have not been completely replaced, but their share in the future value of Waymo's fleet computing power may be under pressure. The focus of competition will shift from merely purchasing peak computing power to achieving system-level efficiency among sensors, computing, power consumption, reliability, and cost for each vehicle.
Source: Public Information
ABAB AI Insight
Waymo originated from Google's autonomous driving project and became an independent company under Alphabet in 2016. Its technical roadmap has long emphasized self-developed full-stack capabilities: autonomous driving software, high-definition maps, simulation systems, LiDAR design, fleet scheduling, and remote operations are all deeply controlled by the company. In 2020, Waymo partnered with Magna to launch its fifth-generation autonomous driving system; the current mass production of its self-developed ASIC indicates that the company is moving its capabilities, previously reliant on general hardware, deeper into the vehicle's electronic architecture, approaching Tesla's use of self-developed chips for FSD hardware and Mobileye's collaborative service model with automakers.
From a capital perspective, the cost of an autonomous driving fleet is not just the one-time chip purchase price but includes the computing power, sensors, electricity, heat dissipation, maintenance, replacement, and downtime costs for each vehicle over its lifecycle. Alphabet can initially bear the costs of ASIC design, validation, and software migration, and once the fleet scales up, it can lower unit operating costs with custom chips. TSMC gains advanced process foundry orders; Waymo focuses on differentiating value in chip architecture, sensor fusion, and collaborative autonomous driving models rather than building its own wafer manufacturing capabilities.
Historically, Apple has integrated hardware, operating systems, and development tools with its A-series and M-series chips, reducing product constraints from general-purpose processors; Tesla, since 2019, has linked neural network inference with onboard power consumption targets after adopting its self-developed FSD chip. Waymo shares similarities in attempting to optimize real-time perception with dedicated computing, but its business model is a high-utilization Robotaxi fleet, not selling driving assistance features to millions of private cars, thus its ASIC returns heavily depend on daily orders and operational mileage per vehicle.
Essentially, this is an industrial chain reconstruction. General chip manufacturers sell standardized computing power, while Robotaxi operators need the lowest reliable operating costs under fixed sensor combinations, software stacks, and long-term fleet conditions. Once the fleet size is large enough, self-developed ASICs can convert part of the value previously paid to general chip suppliers into cost advantages and service profits for operators; however, when the scale is insufficient, R&D, tape-out, and validation costs can make self-development a fixed cost burden. Therefore, chip design rights will concentrate among leading operators as fleet density and capital capabilities increase.
ABAB News · Cognitive Laws
General computing power sells performance, dedicated chips sell cost curves
Self-development is a burden when scale is insufficient, but a rental when scale is sufficient
The ultimate goal of autonomous driving is not in chips, but in cost per mile