中

Beijing Auto Show 2026: Autonomous Driving Finally Moves Beyond PPT

AI, Beijing Auto Show, Autonomous Driving, L3 —

In past years, every time I visited autonomous driving displays at auto shows, I felt like I was watching sci-fi—flashy concepts, but far from production.

But the Beijing Auto Show opening April 24, 2026, is different.

This show is called “the watershed for intelligent manufacturing” for good reason. L3 autonomous driving scaled deployment, domestic chip breakthroughs, AI models on vehicles—things that used to live in PPTs are actually going into production.

Three Signals of L3 Scaled Deployment

First signal: Technology maturity threshold met.

Over the past two years, L2+ assisted driving has proven its safety. Data shows L2+ assisted driving penetration in China exceeded 40% in 2025, with significantly increased user acceptance. This laid the foundation for L3 scaled deployment.

Second signal: Clear policy regulations.

In late 2025, the Ministry of Transport officially released “Management Specifications for Road Testing and Demonstration Application of Intelligent Connected Vehicles,” clarifying liability division and insurance mechanisms for L3 autonomous driving. Policy-level uncertainty is eliminated.

Third signal: Business model validated.

Subscription models are becoming mainstream. Not one-time purchases, but monthly/yearly subscriptions for autonomous driving features. This lowers user decision barriers and gives automakers sustained cash flow.

The Critical Battle for Domestic Chip Breakthrough

Another highlight of this auto show is the collective debut of domestic autonomous driving chips.

Previously, autonomous driving chips were basically monopolized by NVIDIA and Mobileye. But now, Horizon Journey 6, Black Sesame Huashan A2000, Huawei Ascend 610—these domestic chips are approaching or even surpassing international competitors in compute power, power consumption, and cost.

My personal sense is domestic chip breakthrough isn’t a question of “whether” but “when.” And 2026 might be that moment.

AI Models on Vehicles: From Gimmick to Practical

Last year, many automakers promoted “large models on vehicles,” but actual experience showed most just added smart voice assistants that could chat a bit.

This time it’s different.

Multiple automakers demonstrated real large model application scenarios:

  • Multimodal interaction: Not just voice, but understanding gestures, eye movements, expressions
  • Scene understanding: Recognizing complex traffic scenarios, making smarter decisions
  • Personalized learning: Continuously optimizing driving strategies based on user driving habits

These aren’t PPTs; they’re features already validated on production vehicles.

My Take

Don’t get too excited—look at the data calmly.

Autonomous driving from “can run” to “can commercialize” still has many pitfalls. L3 scaled deployment needs to solve at least three problems:

  1. Extreme scenario handling: Current AI models work well in 99% of scenarios, but that remaining 1% of extreme scenarios could be the root of accidents.

  2. Cost control: L3 autonomous driving hardware costs are still in the 20,000-30,000 yuan range. For scale, they need to drop below 10,000 yuan.

  3. User trust: Technology can be great, but if users won’t use it, it’s useless. Building user trust in autonomous driving might be harder than technical breakthroughs.

But at least this auto show shows us autonomous driving is no longer “future tense” but “present continuous.”

Ready to buy your first L3 autonomous vehicle?

Related Posts