Source: Wall Street News

The entry of Tesla's FSD into China seems to be getting closer. Although the timing is uncertain, it is enough to draw buzz and attention from the outside world.

FSD is the core of Tesla's autonomous driving. Prices have been raised many times in recent years. The reason given by the company for the price increase is that Tesla's breakthrough in AI can bring value that matches the price increase.

So what about its features and price? What else is needed to land in China, and what impact will it have?

In fact, Tesla Autopilot is divided into three levels, from low to high, the basic version (Basic), the enhanced version (Enhanced, EAP), and FSD, respectively, corresponding to different functions and prices. Among them, the basic version and the enhanced version are sold domestically. Currently, the FSD is only being launched in the US and Canada. You can pay for the option when purchasing a car domestically, and you can get on the car when FSD is introduced to China.

Since April 2019, FSD has been sold as a standalone feature, and has experienced several price increases since then.

In terms of architecture, FSD has created a full link autonomous driving software and hardware architecture including perception, regulation, and execution at all levels of data, algorithms, and computing power.

Among them, planning: the essence is to solve multi-object associated path planning problems, process the trajectory of oneself and all objects, and guide the car to complete the corresponding execution actions;

Neural network: Control the vehicle by analyzing information such as video streams and outputting the complete kinematic state (position/speed/acceleration/bumps);

Training data: Through 4D automatic labeling technology, upgraded simulation and cloud computing resources, a closed loop of data is formed;

Training infrastructure: including CPUs, GPUs, neural network accelerator units, AI compilers, etc., where the AI compiler can support the new operations required by the neural network and map them to the optimal underlying hardware resources.

AI compilation and inference: that is, running a neural network on a computer. Current inference engines can distribute execution of a single neural network to two independent chip systems.

▌What does FSD need to enter China?

Tesla's August 9 vehicle system update includes the collection of autonomous driving data. Data collection and feasibility verification are carried out starting in a “shadow mode”. Users can independently choose to participate.

Meanwhile, in August, there was media news that Tesla is building its first data center, and its data center business can localize data storage, and all data generated from vehicle sales in the mainland China market will be stored within China. Furthermore, Tesla is setting up a local operation team of about 20 people in China to promote FSD in the Chinese market; and is preparing to set up a data labeling team of hundreds of people in China to provide manual labeling samples for training FSD algorithms.

If FSD were to be implemented in China, data collection, model training, and computing power centers would be regarded as critical.

A June research report by CICC stated that if FSD is promoted in China, the entire data loop needs to be closed domestically, and preparations need to be made in terms of data collection and model training to consider whether the BEV sensing scheme can be well localized and adapted. In China, FSD map surveying and mapping may require cooperation with Chinese manufacturers, and Tesla needs to set up a local computing center and a local algorithm adaptation team.

CITIC Securities also pointed out that, first, map data is highly sensitive and closely related to national security. Exactly how Tesla will obtain relevant qualifications is still unknown, or whether it will conduct joint ventures with Chinese companies, or conduct pilot projects in Shanghai, but it is not easy to roll it out on a large scale in a short period of time.

Second, the pure visual route relies more on extensive data collection and model training for image processing, so Tesla built its own supercomputing center Dojo in the US and developed its own D1 chip to greatly improve FSD training efficiency. The brokerage firm believes that in China, even if Tesla has obtained data collection qualifications, it is likely that it will only be able to conduct model training locally in China. This means that to achieve the same training efficiency as the US, Tesla needs to establish a supercomputing center similar to Dojo in China, which requires a certain development cycle and cost.

▌What is the impact of FSD's entry into China?

Guotai Junan's November 20 research report quoted industry media sources as saying that a blogger certified as a “Tesla China Analyst” on a social networking platform revealed that he had discussed with Tesla the timing of FSD entering China, and Tesla responded that it was necessary to wait for China's relevant autonomous driving laws and regulations to be finalized.

What is the impact of Tesla FSD after entering China? Based on the evaluations of various brokerage firms, it may greatly strengthen consumer awareness, promote the penetration of advanced intelligent driving in China, and trigger a “catfish effect” in the field of intelligent driving in China.

For example, Zhongtai Securities gave an opinion saying that if Tesla opens up the FSD function domestically in the future, on the one hand, it will increase the penetration rate of intelligent driving, and on the other hand, it will stimulate domestic car companies to increase investment in autonomous driving and accelerate the iterative process of domestic advanced autonomous driving technology.

CITIC Securities, on the other hand, said that if Tesla's “extreme cost performance hardware+high margin FSD software” model is implemented on a large scale in China, it may be similar to the “ChatGPT moment” of smart cars; whether at the level of incremental demand or business models, it means new possibilities for current domestic car companies and supply chains.

Analysts further pointed out that car companies that are deeply involved in advanced intelligent driving functions and using this as a differentiated selling point may be expected to benefit, yet since Tesla launched an automobile price war at the beginning of the year, the industry has entered a round of reshuffle and clearance cycle, and low- and middle-end car companies that lack the ability to develop themselves may face greater pressure after FSD enters China.

Editor/jayden