This is the year autonomous-driving chips above 100 TOPS — one TOPS being a trillion operations per second — go into mass production, and the contest around them is heating up: Nvidia’s Orin and Qualcomm’s Snapdragon Ride both reach production this year, and Chinese startups are close behind.
Why compute is the battleground
Automotive is the brightest spot in semiconductors: Gartner puts the 2022 market at USD 65.1 billion, about 12% of global semiconductors and the fastest-growing segment. Within it, autonomy is a compute race — every step from L1 toward L5 raises the requirement. The progression is stark: a 2014-16 Model S ran 0.256 TOPS; a 2019 Model 3, 144 TOPS; the 2022 ET7, 1,016 TOPS. The working forecast: chips above 100 TOPS in production cars by 2022-23, 500 TOPS by 2025, and above 1,000 TOPS mainstream by 2030 — with L3 chips production-ready in 2023 and L4 at scale by 2025.
Why it is hard
Big-compute auto chips face four walls: memory bandwidth that 200+ TOPS designs demand, an AI software stack that must survive algorithms still evolving (hence OTA-updatable, general-purpose architectures), the engineering of very large dies (packaging, power, thermal, yield, cost), and the need for 7nm or 5nm process nodes. Add the automotive specifics: lidar, ultrasonic and millimetre-wave radar and camera pipelines fused on one chip, path planning and control running locally, functional safety, low power — and a 5-to-10-year supply life designed in from the start, because a chip defined today may ship for a decade.
Who is shipping
Mobileye, the incumbent in driver assistance with over 100 million chips shipped, announced EyeQ Ultra at 176 TOPS — sampling to automakers at end-2023, full production 2025. Ambarella’s CV3 claims 500 equivalent TOPS on 5nm, with samples this year. But the ones already at production are Nvidia — Orin at 254 TOPS, entering cars from NIO, IM Motors and Volvo this year, with the 1,000-TOPS Atlan planned for 2025 — and Qualcomm, whose Snapdragon Ride spans 10 to over 700 TOPS and enters production models from Great Wall and GM this year.
The Chinese field
Domestic players converge on the same window. Black Sesame’s A1000 Pro delivers 106 TOPS (INT8); Horizon Robotics’ Journey 5 does 128 TOPS; Cambricon’s upcoming auto chip targets over 200 TOPS on 7nm; Huawei shows full-stack MDC computing platforms rather than single-chip figures. Of the Chinese chips, Black Sesame’s and Horizon’s are the likeliest to reach production cars this year, and the two have argued publicly about what matters — Black Sesame says the chip sets the boundary of what software can ever do; Horizon replies that TOPS is not real performance, and frames-per-second of accurate recognition is the better metric. Both agree on the deadline: the next two to three years decide who gets designed into the cars.
Republished for information. The original text has not been independently verified and does not necessarily reflect our views.
Want to talk through what you just read?
We work end to end, from product design to launch. Tell us about your scenario and we can start there.
Shenzhen XZS Technology Co., Ltd. · Tel 0755-82494862 · 13760424861