Tesla (TESLA) (TSLA.O): Tesla FSD, Breakout Year 2023

We judge that 2023 will be a crucial year for Tesla to break through in the field of autonomous driving. The hardware side will launch the HW4.0 version, and the software side has tested the BetaV11 version in a small range, and the feature richness is gradually improving. The regions with a high FSD penetration rate of the company are mainly in North America. At present, the United States has issued corresponding documents at the national level, and various states have also published multiple papers, one after another. The overall attitude is supportive. Relying on the systemic advantages brought by the integration of chips + algorithms + data and the cost advantages of pure vision solutions, stability & scalability, Tesla is expected to enable Tesla to drive autonomously. The technology keeps leading in the global market and constantly strengthens its position in the global electric vehicle market. We remain optimistic about the company's competitive advantages in this field and the company's medium and long-term investment value.

1) Autopilot is the basis for Tesla to break away from the traditional vehicle logic and realize the vehicle + data two-wheel drive business model. The current market has many differences and concerns about the progress of FSD and the monetization space. From the hardware and software perspective, 2023 will undoubtedly be a pivotal year for Tesla's autonomous driving changes.

2) From the perspective of Tesla's development history: After the FSD self-developed chip was loaded into the car, Tesla's autopilot technology accelerated iteration: whether it is data collection and processing, or perception/regulation, it has gradually established profound barriers. Therefore, an in-depth analysis of Tesla's FSD system is the key to understanding Tesla's core competitiveness, which is also the core focus of this report.

1. Hardware: from outsourcing to self-developed, the computing power and performance have been greatly improved; pay attention to the follow-up progress of HW4.0

1) HW1.0 adopts the millimeter-wave radar cooperated with Bosch, EyeQ3 chip and NVIDIA Tegra3 cooperated with Mobileye. Mobileye mainly provides algorithms. 2) In the HW2.0 era, Tesla cooperated with Nvidia and adopted its DrivePX2 chip, greatly increasing the number of sensors and cameras. 3) Compared with the HW2.0 version, the biggest breakthrough of HW3.0 is that it is equipped with a self-developed unmanned driving chip FSD, which gradually switches from external supply to self-developed fields. 4) HW4.0 will usher in a new upgrade and is currently under normal research and development. The company expects to be mass-produced on Cybertruck in 23H2.

(a) According to Zuosi Auto Research, HW4.0 internal CPU cores increased from 12 to 20, with a maximum frequency of 2.35GHZ. The number of TRIP cores rose from 2 to 3 with a maximum of 2.2GHz (currently HW3.0 The central frequency of the chip is 2GHZ). At the same time, HW4.0 also adopts the same dual FSD chip redundancy design as HW3.0. Regarding functions, HW4.0 will have the tasks of driving on roads without map data and emergency vehicle detection, optimizing the identification of vulnerable groups, and increasing the recognition of other vehicles' turn signals.

(b) According to Musk's 22Q4 performance exchange meeting, HW4.0 will first be used in the Cybertruck field, which is expected to start mass production and delivery in 23H2.

2. Software: The A.P. system has a long history of development. It has been updated to the BetaV11.3 version. The richness of functions has been continuously improved, and the company's self-developed barriers have gradually emerged.

1) The FSD software has been updated to BetaV11.3. (a) In January 2023, Musk tweeted that the company plans to release FSDBeta V11.3. With this release, Tesla will use neural networks for navigation and control, not just vision. (b) From the perspective of the richness of functions, the BetaV11 version has improved the occupancy network accuracy, reduced the error rate of vehicle parking judgment and other indicators, and improved the level of intelligence.

2) From improved intelligence levels, the first edition came out in 2014; Tesla AP has been continuously updated and iterated and has achieved remarkable self-research results in many fields, including simulation, perception, planning, and data, building Tesla's autonomous driving. The moat of the solution includes explicitly: (a) In terms of simulation, Tesla's simulation technology now has the characteristics of rapid labeling, reproducibility, and extreme long-tail scene simulation. As of AIDay 2022, Tesla announced that it had drawn 2,000+ kilometers of the road environment, obtained 3.71 billion pictures and 480 million annotations through simulation, and integrated them into the car model to improve FSD performance. (b) Regarding data, "Shadow + Fleet" creates a unique aRegardingRegardingt closed-loop solution, which is far superior to other car companies represented by Waymo. The closed loop of data accumulation + training constitutes one of Tesla's core competitiveness in algorithms. (c) Regarding data labeling, Tesla adopts the method of man-machine cooperation labeling, which significantly improves the efficiency of data labeling. (d) In terms of perception, the algorithm iterates rapidly, evolving from two-dimensional to multi-dimensional: Occupy network algorithms are expected to solve long-tail problems and emergencies, laying a solid foundation for pure visual paths. The proposal of occupying the network algorithm can make accurate estimates of the movement patterns of different bus parts at the same time point after model parts effectively solve the problem of perception of long-tail obstacles. (e) Regarding regulation and control, the hybrid planning system + joint planning is an effective solution for the non-convex behavior space and multidimensionality of parameters in autonomous driving.

3. Commercialization: The current penetration rate is about 7.4%. In the future, we will pay attention to releasing new models. The overall U.S. policy is supportive

1) TroyTeslike data shows that the global penetration rate of Tesla FSD in 22Q3 is about 7.4%, about 28.3pcts lower than the penetration rate of 35.7% in the peak period of 2019Q4. The main reason is that (a) the FSD version has not been significantly updated; (b) the proportion of sales of models with lower prices, such as Model3/ModelY, continues to increase.

2) Looking forward to the future, we believe the launch of Tesla Semi and TeslaBot will vastly accelerate the largelyentially's autonomous driving software. Among them, Semi relies on the three-team arrangement and self-driving technology to expand to the TeslaBot (humanoid robot) field, which can accelerate the company's commercialization progress in the field of self-driving to a large extent.

3) The U.S. policy is generally supportive. (a) From a national perspective: the National Highway Traffic Safety Administration (NHTSA) and the U.S. Department of Transportation released the "Federal Autonomous Vehicle Policy," and the U.S. Department of Transportation formulated the Automated Vehicles Comprehensive Plan (AutomatedVehiclesComprehensivePlan). and presented a vision for the form of future transportation. (b) From the perspective of state governments, as of January 2023, 29 states and the District of Columbia have passed legislation, ten state governors have issued executive orders, and nine state laws are on the ballot or hold. States only a minimal number of states have not taken any action for the time being.

Risk factors: the risk of further deterioration of the global epidemic; the risk of intensified international trade conflicts; the risk of valuation fluctuations caused by severe safety accidents in self-driving cars; the chance that the implementation of self-driving policies is less than expected; the development of artificial intelligence technology is lower than expected, etc.; the event that the progress of the company's Berlin & Austin factories will not meet expectations, etc.; the risk of insufficient production capacity of the company's power battery suppliers; the risk of inadequate power grid load, etc.

Investment suggestion: As a leader in the global electric vehicle field, we continue to be optimistic about Tesla's leading advantages in battery, BMS, Autopilot algorithm, and intelligent driving data accumulation. In the future, the penetration rate of high-margin businesses such as FSD & vehicle insurance services will continue to increase. , the company's profitability is also expected to improve further. We maintain the company's 2023-2025 revenue forecasts at US$1030/1402/180.7 billion and keep the company's 2023-2025 Non-GAAP net profit forecasts at US$142/214/292 billion, optimistic about the company's short-term, mid-and long-term investments value.