What Metrics Matter in the Autonomous Driving Shakeout?

Let me share something. Changan Vice President He Gang judges that the smart driving industry will complete its first round of reshuffling around 2028, leaving possibly only five or six competitive solution providers in the Chinese market, adding, "We and Huawei's Yinwang will survive." This statement has momentum, but my first reaction was that smart driving demos are still far from mass production.

From my perspective, smart driving looks very similar to embodied intelligence. Running in videos doesn't equal deliverability. The real hurdles for solution providers are stable vehicle installation volume, yield rates, failure rates, automotive-grade validation, cost, and after-sales responsibility. Recently engaging with solutions like Minieye, what I care about most is abnormal takeover, interface stability, and who takes responsibility when problems occur. Changan has both self-research and cooperation legs, and has obtained L3 pilot approval, which counts as hard evidence. But between pilots and batch installations lies an entire supply chain.

Reshuffling will wash away companies without core metrics. Startups shouldn't bet on "whether we'll survive," but first ask themselves if they have verifiable deliveries. By 2028, surviving companies must speak through delivery records.