“Science and Technology Innovation Board Daily”, September 5 (Editor Song Ziqiao) Today, Tesla's AI team released a product roadmap. Among them, it is expected that fully autonomous driving (FSD) will be launched in China and Europe in the first quarter of 2025, but it is still awaiting regulatory approval.

Since 2016, Musk has been exploring Tesla's FSD autonomous driving solution. In 2024, Tesla FSD V12 was launched and launched on a large scale in North America. FSD V12 autonomous driving was upgraded to an “end-to-end” neural network from perception to vehicle control, and training computational power+data became the core barrier.

Currently, FSD iteration is progressing steadily. The parameters of the latest 12.5.2 version have been raised to 5 times that of 12.4, and the highway and city technology stacks have been merged to further improve driving fluency. Specific improvements include more natural and earlier lane change decisions, driver monitoring using only a vision system, end-to-end driving on motorways, and FSD functions on Cybertruck.

At the same time, Tesla continues to invest in the construction of AI supercomputing centers to meet its computing power requirements for autonomous driving training. As of 24Q2, the AI computing power of Tesla's self-built supercomputing center exceeded 0.04 million, the equivalent computing power of Nvidia's H100, and is expected to rise to close to 0.09 million by the end of the year.

Musk previously predicted that Tesla's FSD test mileage would need to reach 6 billion miles to meet the requirements of global regulators. According to Tesla's latest earnings report, as of the end of Q2, the FSD's cumulative mileage exceeded 1.6 billion miles. The company also said that by the end of the year, FSD's uninterfered driving mileage will exceed that of human drivers, and it is confident that it will achieve the level of unsupervised L5 smart driving next year.

Since this year, Tesla has lowered FSD offers several times to attract users. China is an important market for Tesla cars, but FSD's entry into China is constantly being delayed.

Recently, Musk said during Tesla's earnings call in July that FSD is expected to enter China by the end of the year, FSD will land in China, Europe, and other countries after the 12.6 version is released, and the unsupervised version is expected to be commercialized at the same time as Robotaxi.

At the time, he said that the promotion of Tesla's autonomous driving technology in the US and other regions will not encounter regulatory barriers.

Although the Tesla FSD has not yet been launched in the Chinese market, it was already sold with the car as early as the Model 3 went on sale, and a subscription system may also be introduced in the future.

Guoxin Securities said earlier that in the long run, Tesla is expected to build a series of artificial intelligence terminal carriers (smart driving cars, robots, etc.) to enable new profit models through “hardware cost reduction, software bonuses, and Robotaxi shared travel.” The software uses a combination of FSD software, software stores, and subscription services to increase added value and open up intelligent driving application scenarios for shared travel.

Ping An Securities further stated that FSD's entry into China may become a “catfish” in the domestic autonomous driving field, speeding up the implementation process of advanced smart driving in China, and leading car companies or suppliers in the field of advanced smart driving will benefit even more.

Learn about Tesla FSD

Tesla FSD iterated from the initial version to V12, and its technical innovation mainly lies in the restructuring of the architecture — the introduction of an end-to-end architecture. Meanwhile, the continuous iteration of Tesla's FSD system marks that its autonomous driving technology is entering a period of rapid data-driven development.

The advantages of Tesla FSD are reflected in: 1. Algorithms: deep integration of machine learning in the entire autonomous driving process; 2. Perception: it can perform more driving actions that were difficult to complete before, such as planning routes based on real-time traffic conditions rather than only relying on maps; 3. Computing power: Tesla also plans to increase the number of GPUs by two to three times as much as using thousands or even tens of thousands of high-performance H100 video cards. This move will greatly improve data processing capabilities and training.

Currently, Tesla FSD is implementing a series of promotional policies in the US market. According to statistics from Tesla's software tracking website, as of April 26, of the 19759 Tesla vehicles registered by its paying users, 7,444 had the FSD V12 version installed, and 9724 had the FSD V11 version. Among them, the largest number of FSD V12.3.4 and FSDV11.4.9 were 6149 and 9381 vehicles, respectively, with penetration rates of 31.1% and 47.5%, respectively.