Tesla’s “FSD” in China is pretty good, but there is a catch

Whereas in the U.S., Tesla is all alone in this department of advanced levels of automated driving, it’s much more cutthroat in China, where multiple brands already offer smart driving capabilities like, if not better than, FSD (Supervised). Huawei, Li Auto and Xpeng (maybe Xiaomi) right now are the front runners when it comes to offering the so-called all-scenario Navigate on Autopilot (NOA) features covering both highway and urban road conditions. The race right now among these pack leaders is to push out the so-called parking space to parking space automated driving features covering all road conditions, minimizing driver intervention, but still Level 2, like what the latest FSD Supervised can do in the U.S.

“NOA available anywhere you drive, day or night” right now is the game’s name.

I have watched some of the live streams, which have been happening day and night for the last couple of days and stressing the system, e.g., going through tight Hutongs (a type of narrow street or alley commonly associated with northern Chinese cities) in Beijing or going through the so-called “city villages” in some of the southern cities like Guangzhou/Shenzhen.

Here are some of my takeaways:

Most of the media/KOLs that have done the tests have been impressed with the capabilities of this initial iteration, given Tesla’s data collection/transfer limitations in China. The system performs especially well in easier scenarios like highways and ring roads where traffic is less complex. Also, it is quite impressive in certain urban scenarios, but the system is quite conservative in yielding to pedestrians on a crosswalk or to parked/incoming vehicles.

Where the system lacks is the understanding of local traffic rules (such as no use of shoulder/bike lanes on turns, similar to the bus lane rules that Elon talked about in the most recent earnings call) and the sporadic use of wrong lanes (e.g. going straight in a left or right turn only lane) or navigation showing the vehicle in one lane when in fact it’s in another or wrong perception of objects (red balloons as traffic lights). Many of the live streams counted the number of traffic violations from the vehicle and the number of points that would have been taken off or licenses suspended (12 points = suspension) as a result.

The consensus seems to be that this initial version of “FSD” does not catapult Tesla into the first camp of smart driving leaders. At this point, the “FSD” China capabilities are nowhere near the latest version in the U.S., but impressive nonetheless, given Tesla’s data collection limitations in China.

But here’s the catch: of the hundreds of thousands of Tesla owners in China, only a few thousand of those owners, to the best of my knowledge, have access to the initial version of the “FSD.” A reminder that to gain access, an owner must have their vehicle running on HW4.0 and have purchased the “FSD” software package (V13.2.6), which costs RMB64,000 or $8,800 in China. And, the system must be running on the latest Baidu map software, which is included in the software update. Many Tesla owners in China bought FSD as far back as 2019 on HW3.0, which means they won’t be able to use it, at least in its current format.

So going forward, the dilemma for Tesla is figuring out two things to democratize this feature (if this is its aim) for more customers in China (a more significant share of them has HW3.0):

1. Subscription model (RMB64,000 upfront cost is too high given competition) or discounted variations;

2. Make it available for HW3.0 (which I’m hearing is being planned).

On a bigger scheme of things, this “FSD” update in China only qualifies Tesla to have a seat at the table among the leaders of the pack; it doesn’t separate itself from the rest of the pack if we are talking about all-scenario L2++ automated driving (and democratizing it, which Tesla is still behind). This competitive track in 2025 will be white hot with more players joining (e.g. BYD God’s Eye/DiPilot, Mercedes L2++, Zhuoyu or DJI Auto, Momenta, DeepRoute, Horizon Robotics SuperDrive, etc.).

A small win for FSD but a bigger win for smart driving in China.

It will be interesting to see how fast “FSD” capabilities in China can be iterated from now on, given the limited number of vehicles that will actually have access to the initial and later iterations of the feature and what Tesla will do (and sacrifice) to democratize the feature in China, if at all.

After all, democratizing and standardizing higher levels of automated driving was the message from arch-rival BYD earlier this month when it rolled out the base versions of God’s Eye assisted driving systems to 21 models across its Dynasty and Ocean network models, FREE OF CHARGE. The same message has come from He Xiaopeng and Li Xiang, among others, as they up the ante in their own smart driving game and openly invite the media to compare test Tesla’s “FSD” with advanced NOA capabilities of their own vehicles, which they believe are much more superior.

Tesla is facing immense pressure to grow volume this year, given the recent political backlash from Elon Musk and the general slowdown in EV uptake in various markets worldwide. China became even more crucial as local deliveries of the new Model Y began yesterday, with the cheaper “Model Q” on the horizon for China and the rest of the world.

Getting FSD tactically right and making it available to more customers in China despite the challenges of data restrictions could go a long way toward helping Tesla maintain strong volumes in its largest and most crucial market.