---
title: "China's self-driving brains just hit 10 million — and the hard part is next"
date: 2026-10-04
category: Chips & Compute
site: NeuroAI
canonical: https://neuroai.site/a/na-chips-automotive-ai-horizon
language: en
---

# China's self-driving brains just hit 10 million — and the hard part is next

> Horizon Robotics passed 10 million shipped Journey (征程) automotive AI chips by mid-2025, but the real test is moving from assisted driving in premium cars to the mass market.

The chip that decides whether your car brakes for a child is no longer a lab experiment in China.

It is now a mass-produced part, sitting in millions of vehicles, built by a company most Western drivers have never heard of.

The milestone crossed this summer was not a press-release number — it was a manufacturing one.

## The 10-million mark

Horizon Robotics (地平线) announced in late August 2025 that its **Journey (征程)** series of automotive AI chips had passed **10 million units** in cumulative mass-produced shipments. The company says the chips now power more than **400 vehicle models** from over **40 automakers**, serving upwards of **6 million car owners**.

That scale makes Horizon the first Chinese intelligent-driving compute supplier to clear the ten-million threshold. For context on why that matters: industry trackers put advanced driver-assistance systems (ADAS) in well over half of new Chinese passenger vehicles, and analysts estimate foreign suppliers still dominate the high-end autonomous-driving chip segment. A domestic player reaching ten million units is the clearest sign yet that "made-in-China" no longer means "entry-level only" for car brains.

## What the Journey 6 actually does

The current generation, **Journey 6**, entered mass production in early 2025 and spans a wide compute range — from **10 to 560 TOPS** — built on Horizon's third-generation **BPU (Nash)** architecture. The family is deliberately sliced into tiers:

- **Journey 6B** — entry-level active safety; Bosch is building a camera platform on it, with volume production targeted for mid-2026.

- **Journey 6E** — about **80 TOPS**, pitched at the cost-sensitive mainstream.

- **Journey 6M** — about **128 TOPS**, for city-assist driving with multi-sensor fusion.

- **Journey 6P** — the flagship at **560 TOPS**, more than four times its Journey 5 predecessor, anchoring Horizon's full-scene **HSD (Horizon SuperDrive)** system.

The strategy is "智驾平权" — making assisted driving affordable across the board. BYD has shipped over a million vehicles on earlier Journey generations and, in 2025, pushed a Journey 6M-based system down into the roughly 100,000-yuan car class. Changan, Geely, GAC, and Chery have also adopted Horizon solutions.

## Why this is an AI-chip story, not a car story

Each Journey chip runs a purpose-built **BPU (智能驾驶专用计算架构)** — a domain-specific AI accelerator tuned for the transformer-heavy perception and planning stacks modern ADAS demands. Horizon says a decade and three BPU generations lifted compute performance more than **1,000x**. The company sells not just silicon but four engagement models, from licensing the BPU architecture so automakers design their own chips, to turnkey systems that bundle chip, OS, and driving software.

Global Tier-1 suppliers have noticed. Bosch and ZF have integrated Journey 6, and Volkswagen is among the automakers using Horizon silicon for China-market vehicles, with overseas production of Journey 6E/6M-based systems expected to begin in early 2026.

## The bottleneck is no longer the chip

Horizon's own numbers reveal the next problem. Its average per-vehicle value rose about **70%** year-on-year as buyers moved from basic L2 to city-navigation assist — but the company itself flags that the industry must balance technological lead, scaled production, and profitability. The economics of automotive AI are brutal: a chip good enough for a 300,000-yuan SUV must also make sense at 120,000 yuan, where margins vanish.

There is also the autonomy question. A 560-TOPS chip running "China's FSD"-style software is impressive; proving it is safer and smoother than a human across millions of edge cases is the multi-year grind no amount of TOPS solves.

## What readers can do now

- If you buy a Chinese-brand car in 2026, check whether the assist system runs on a domestic chip like Journey 6 — it increasingly affects both capability and repair cost.

- If you build or source automotive electronics, evaluate Horizon's IP-licensing model as an alternative to buying a black-box chip, especially for cost-sensitive lines.

- If you follow the sector, watch the 150,000-yuan price band: that is where domestic ADAS chips either prove mass-market viable or stall.

## Honest limitations

Figures on the 10-million shipment milestone, model counts, and TOPS tiers come from Horizon's official statements and Chinese financial press (China Securities, China Quality News), cross-checked against DigiTimes reporting. The "foreign suppliers dominate ~70% of high-end ADAS chips" figure is an analyst estimate from a single outlet and should be treated as directional, not precise. I have not independently verified Horizon's 1,000x BPU performance claim or its per-vehicle-value growth against audited filings, and I have excluded any stock-price or investment recommendation — this is a technology and market-structure analysis, not investment advice. Safety and real-world reliability of HSD were not independently tested here.

---

Published by NeuroAI (https://neuroai.site/) — https://neuroai.site/a/na-chips-automotive-ai-horizon
Free to quote with attribution and a link to the original.
