200 Robotaxis going overseas: don't just look at the numbers

Let me share something. Pony.ai and FutureLink plan to deploy the first batch of 200 vehicles in South Korea, with certification expected around 2028. My first instinct was to look at the sample size and timeline: 200 vehicles is too few, and 2028 is too long.

The plan is to transition technical validation previously conducted in the Gangnam autonomous driving demonstration zone in Seoul into the commercial operation phase.

This sentence is more critical than the number 200. It indicates that the current focus is still on migrating from demonstration zone validation to actual operations.

Demonstration zone data is like backtesting samples—beautiful but potentially biased. Real operations must deal with certification, road long-tails, liability boundaries, and user acceptance.

I've built price elasticity models for years, and what I fear most is linear extrapolation. When we discussed WeRide before, I mentioned that high growth rates are often base-effect traps. The same applies here. If those 200 vehicles are concentrated in Gangnam, Seoul, it's merely an observation window; confirming scalable revenue is still far off.

The risks are also clear. Completing certification around 2028 means policy milestones are still distant. Adapting algorithms to Korean roads requires much more than just swapping maps. Rain, night driving, narrow roads, and non-motorized traffic flows will cause takeover rates to spike. I used a pure vision solution for three weeks and know that as edge cases multiply, model stability needs re-evaluation. In live trading, you consider slippage; commercialization is no different.

Going forward, I'll monitor certification progress, fleet expansion, remote takeover rates, and unit economics models. The initial batch of 200 vehicles alone isn't enough to support the judgment that commercialization is already established.