The chip that helps a car see the road used to be a costly option reserved for flagship models. That assumption is being rewritten in China, where assisted-driving compute is moving down the price ladder fast. One local supplier has now shipped enough of these chips to cross a once-distant milestone.
Horizon Robotics (地平线) builds the silicon brains behind assisted driving (智能驾驶) in a growing share of Chinese vehicles. In August 2025 the company said its Journey (征程) family of automotive chips had passed ten million units shipped cumulatively — the first Chinese supplier of this kind to reach that scale.
From safety chips to city driving
Horizon's story is really a story about how fast the software-defined car absorbed more compute (算力). Its early chips handled modest active-safety tasks. The latest Journey 6 generation, built on the company's third-generation BPU "Nash" (纳什) architecture, is designed to span the whole range of assisted driving, from basic safety to full city navigation.
The lineup is deliberately wide:
- Journey 6B: entry-level active safety and parking, the version Bosch chose for a new camera platform.
- Journey 6E / 6M: mid-tier driving at 80 TOPS, the workhorse for mass-market models.
- Journey 6P: the flagship at 560 TOPS, powering Horizon's own SuperDrive city-driving system.
By Horizon's account the series covers 10 to 560 TOPS, letting a carmaker pick a chip sized to the price of the vehicle rather than over-provisioning every model. That flexibility is the core of its commercial pitch.
What actually shipped
The ten-million figure is not just a logo on a slide. Horizon says the Journey 6E entered mass production and delivery in August 2025, with early models including the Aion "Bawanglong," the MG4, the Roewe M7 DMH, and several Chery vehicles. The higher-end 6M and 6E together had gathered more than 20 OEM platform commitments covering over 100 vehicle nameplates, and the broader Journey 6 family had drawn interest from more than 40 brands across more than 400 models.
The flagship 6P, rated at 560 TOPS, had already returned from the fab and completed live-vehicle validation within 100 hours, according to the company — a sign of how compressed automotive silicon timelines have become. It underpins Horizon SuperDrive, a city-assist system the company is positioning against the best global offerings.
The business behind the mileage
Behind the unit counts is a real company with real revenue. Horizon reported 2024 revenue of about ¥2.38 billion (about US$336 million / HK$2.62 billion), up roughly 54% year on year, and said it held about a third of the domestic-brand passenger-car assisted-driving compute market that year — a claim the company itself makes, and one worth treating as directional rather than independently audited.
The strategic move is to sell not just chips but a full stack: silicon plus software plus a development environment, so that a carmaker can adopt assisted driving without building the expertise from scratch. That "maximum common denominator" approach is why Horizon describes itself as the Wintel of the robotaxi-era car, a bold analogy that captures its ambition more than its current reach.
Why this matters beyond China
Two things make the Horizon case interesting for global readers. First, it shows that the center of gravity for automotive compute is shifting toward suppliers who can pair cheap, scalable silicon with fast software delivery — exactly the kind of embodied intelligence (具身智能) pipeline that robots and vehicles share. Second, it is one of the few Chinese chip designers to list publicly outside the mainland, with a Hong Kong debut in 2024 that was the exchange's largest technology offering that year.
The large model (大模型) wave also helps: transformer-based perception runs far better on the new BPU than on the previous generation, and Horizon says transformer efficiency improved roughly tenfold on the 6E, which is what lets mid-priced cars run capable assisted driving without a costly cooling and power budget.
Trend reading
According to Li An, chief scientist at BrainNet (脑机网), the car is becoming the largest volume deployment of edge intelligence in the world, and the supplier that wins there sets the template for robots and other moving machines.
Honest limitations
The shipment and market-share figures are company-disclosed and widely reported by Chinese auto and financial media, but the market-share claim in particular is Horizon's own and is not independently audited, so treat "about a third of the market" as the company's view. TOPS ratings are vendor specifications and say little about real-world driving quality, which depends on software, sensors, and validation. Automotive silicon is also cyclical and competitive: global GPU leaders and other domestic chipmakers are all targeting the same cockpit, and a single failed model program or a shift in subsidy and safety regulation could reset demand quickly.
What readers can do now
- Investors: judge Horizon by recurring design wins and SuperDrive adoption, not just cumulative shipments, since a chip win must convert into multi-year production volume to matter.
- Engineers: if you build vehicle software, benchmark the full stack (chip plus driving software) rather than TOPS alone, because integration quality decides the on-road result.
- Buyers: when shopping for a car, ask which assisted-driving compute generation it uses and whether city-driving is shipped or promised, since 6E/6M versus 6P implies very different capability.