Let Tesla FSD Enter China, Musk is More Eager Than Anyone
Tesla’s FSD Enters China but Is Not Fully Integrated Yet.
On the evening of February 24, Bloomberg reported that Tesla is preparing a software update for Chinese customers, which will soon deploy the FSD (Full Self-Driving, priced at 64,000 yuan) feature in China. The next day, Tesla China announced that it has started a phased rollout of the 2024.45.32.12 software update, upgrading the city roads’ Autopilot for users who ordered the FSD intelligent assist driving feature.
Both pieces of information mention the upgrade of functions such as “recognizing traffic signals, making turns, handling changes in lanes and speed, and automatically changing lanes based on speed and route,” making “Tesla FSD entering China” the center of discussion.
Compared to the FSD capabilities in the U.S., this software update from Tesla cannot achieve full self-driving but is merely an optimization of the EAP feature (Enhanced AutoPilot, priced at 32,000 yuan).
In other words, although everyone is talking about the FSD feature going online, what is actually being launched is a “trimmed-down version” rather than the fully autonomous and ten times safer than human drivers’ FSD feature that Musk has touted.
The complete version of FSD entering the Chinese market still faces many uncertainties.
What Are the Uncertainties?
“Don’t expect too much without Chinese data.” A senior executive from an intelligent driving solution provider bluntly stated about the already deployed FSD feature. The inability to collect on-the-ground road data for training is a bottleneck preventing Tesla’s FSD from truly landing in China.
FSD is facing a dilemma in terms of data compliance between China and the U.S. “China does not allow Tesla to transfer training data overseas, while the U.S. government limits Tesla’s training in China,” Musk said in the 2024 fourth-quarter earnings call. This has nearly reached an impasse.
Since adopting the “end-to-end” technology architecture (inputting data obtained from sensors at one end and directly outputting control results at the other), data-driven approaches have become particularly crucial, requiring large amounts of actual driving data to train the models, so they can actively learn human driving habits.
However, the problem is that Tesla currently finds it challenging to collect road data from China, and until compliance work is in place, it can only use engineering vehicles to gather data in limited areas. This means that even if it sells nearly 660,000 cars in China annually (2024 sales in the Chinese market), the amount of local data it collects is still a concern.
Musk stated that, at this stage, the FSD must be trained on Chinese road videos found online, but this data quality is hardly comparable to on-the-ground data collection. Moreover, Chinese traffic regulations and road environments differ significantly from those in the U.S., with special areas like bus-only lanes, guidance islands, and waiting areas being unique scenarios in China.
Musk previously revealed in an earnings call that one of the biggest challenges for FSD in China is bus-only lanes. “In China, there are several hours in a day dedicated to bus lane usage, and if you accidentally enter a bus lane at the wrong time, you’ll receive a ticket right away,” Musk said.
Difficulty in understanding traffic regulations is one aspect, but the presence of couriers, delivery riders, and other traffic participants in China further increases the complexity and demands higher safety standards from FSD, which all require extensive data support for its training.
To promote FSD deployment in compliance with local regulations, Tesla has been seeking to bypass some sensitive issues, for example, by deploying data centers in China or switching its map data service provider to Baidu Maps. However, to meet regulatory requirements in both China and the U.S., Tesla needs to find more solutions.
In addition, computing power is a major bottleneck. Tesla built the Cortex supercomputer cluster overseas to train FSD, with about 100,000 NVIDIA H100 and H200 chips, making it the car company with the largest computing power scale, but this cannot be used for domestic data training.
Due to international policy impacts, Tesla is also attempting to resolve the computation power bottleneck in China; there are reports that it may lease or build a computing center in China to train FSD, but whether this training center’s computational power can meet the vast data training needs in China remains to be seen.
Who Is Anxious About FSD Entering China?
Even if it’s not the complete FSD form, Tesla is anxious to push it first.
Tesla’s sales have hit a growth bottleneck. Financial reports show that Tesla delivered 1.789 million new vehicles in 2024, a 1.1% decrease compared to 2023 (1.8086 million), marking the first decline in new car sales for Tesla in a decade. The sales volume of two bestselling models, the Model 3 and Y, has shown signs of fatigue.
Tesla’s delivery performance over the past five years saw a decline in 2024.
In the Chinese automobile market, there are numerous models targeting the market segment of Model 3 and Y. Last year’s hit model Xiaomi SU7 entered Model 3’s market range, and Model Y has been facing competitors from all sides since the second half of last year, including Le Tao L60, Zhi Ji LS6, Avita 07, Zeekr 7X, and Zhi Jie R7, all directly targeting Model Y’s price range.
The increased competition in the market has weakened Tesla’s product competitiveness. Additionally, Tesla has been relying on the Model 3 and Y for nearly ten years (Model 3 launched in 2016, a redesign in 2023; Model Y released in 2019, redesigned scheduled for January 2025), and the constant delay of lower-priced models like Model 2/Q has further impacted delivery performance.
Tesla’s delivery struggles will become more severe. Since the beginning of the year, several domestic automakers and suppliers, led by companies like BYD and Changan, are promoting “intelligent driving equality,” with advanced intelligent driving capabilities (featuring highway NOA, automatic parking, and partial city NOA functionalities) gradually being introduced to models priced around 80,000 yuan.
When advanced intelligent driving gradually becomes standard in domestic models, Tesla’s requirement to pay an additional 64,000 yuan to buy the intelligent driving package will further erode its appeal, and the non-advanced intelligent driving capabilities of the Model 3 and Y will face diminishing competitiveness.
In the Chinese market, Tesla urgently needs FSD to bolster its position. Even if its a trimmed-down version, Tesla has to compromise first to avoid missing out on market opportunities; after all, the Greater China region has always been a core source of sales for Tesla.
In contrast, domestic automakers are not overly concerned about FSD entering China. Public statements show that automakers’ attitudes generally fall into two categories: support for its introduction to leverage Tesla’s influence for intelligent driving promotion and education; or waiting for Tesla’s entry to compete in domestic road scenarios.
He Xiaopeng has publicly stated that if Tesla’s FSD technology enters China, it will have a positive impact on intelligent driving companies, including XPeng and Huawei.
Domestic automakers have yet to compete with FSD on the same platform, and the entry of the trimmed-down FSD version mainly serves as a response to users who have already purchased the FSD intelligent assist driving feature, but this minority of users is unlikely to make much of an impact in the market competition.
At the end of the day, this is a gamble by Tesla. Although “FSD entering China” is a hot topic, rushing to launch an immature FSD version in the Chinese market might ultimately affect Tesla’s reputation.
Editor: Zhongxiaowen
