TESLA(TSLA.O)：3Q DELIVERIES MISSED DUE TO LOGISTICS DISRUPTION AND ROBOT PROGRESS GIVES NO SURPRISE

　　Core views:
　　Tesla held its annual artificial intelligence day on Oct 1 Beijing Time in Palo Alto, California. The AI Day 2022 focuses on updating the technical research and engineering progress of the three major projects, namely, the humanoid robot Optimus, the Full Self Driving (FSD) system and the supercomputer Dojo. At the same time, Tesla announced its 3Q22 production and sales figures, which were lower than market expectations. But that's mostly dragged down by logistics hiccups, not demand. The supercomputer Dojo is specially designed for deep neural network training and will provide computing power for Tesla FSD model training. Tesla expects to launch the ExaPOD supercomputer in 1Q23. It also showed off a prototype of the Optimus robot, which performed simple actions like walking and moving objects. In core technology links such as identification and planning, Optimus is expected to share most of the technology with Tesla Autopilot. In terms of autonomous driving, the number of testers of the FSD Beta version has grown rapidly, and the Company expects to launch it globally from North America by the end of 2022. Due to a number of setbacks including the Fed’s continued tightening expectations, disappointing 3Q22 vehicle deliveries, little surprises from the AI Day 2022, and Elon Musk’s restart of the Twitter acquisition, Tesla's stock price plunged recently. However, we consider these setbacks more of a temporary disruption, and the current market concerns are obviously overdone. In the medium to long term, Tesla's combined advantages of “automobile + software” could continue to support its rapid and highly-visible growth. We advise investors to actively focus on opportunities following its recent share price correction.
　　Abstract:
　　Automobile business: 3Q22 deliveries missed expectations due to logistics hiccups rather than demand weaknesses. Tesla announced its 3Q22 global electric vehicle (EV) production and sales data. 1) Deliveries: The Company delivered a total of 344,000 EVs in 3Q22 (+42% YoY, +35% QoQ), lower than market expectations of 363,000 vehicles. This includes 325,000 Model 3/Y (+40% YoY, +36% QoQ) and 19,000 Model S/X (+101% YoY, +16% QoQ). 2) Production: The Company produced a total of 366,000 EVs in 3Q22 (+54% YoY, +42% QoQ), including 346,000 Model 3/Y (+51% YoY, +43% MoM), and 20,000 Model S/X (+123% YoY, +21% QoQ). 3) Influencing factors: In the past, Tesla adopted mass production methods for different regional markets, and the delivery volume often increased significantly at the end of the quarter, which often caused the logistics cost to increase during the peak delivery season. In this regard, Tesla adjusted its production plan in 3Q22, and the proportion of weekly production for different regional markets is more even. This has resulted in an increase in the number of vehicles in transit, which will be delivered upon arrival at their destinations.
　　AI: Tesla held the AI Day 2022 event and announced the R&D progress of a number of products. Tesla showcased the technology development and engineering process of the humanoid robot Optimus, the autonomous driving solution FSD, and the Dojo supercomputer. In our view, Tesla has a clear master plan and development route for supercomputers (for autonomous driving model training), autonomous driving (for robot and car perception/decision-making/motion), smart cars and humanoid robots in the medium to long term. We remain optimistic about its “automobile + software” solution.
　　The humanoid robot Optimus: Likely to be mass-produced in 3-5 years, and the long-term target price is less than US$20,000. At the event, Tesla showcased a prototype of a humanoid robot. 1) Appearance parameters: The appearance is similar to that of an adult man, the whole body has 200+ degrees of freedom, the weight is 73kg, the torso is equipped with 28 joints, and the palm part is equipped with 6 brakes to complete 11 free movements; 2) Technology deployment: Optimus will run on Tesla SoC and support Wifi and LTE connections. Optimus shares vision algorithms and computing chips with Tesla, and uses a 2.3kWh and 52-volt battery pack. For different actions, the R&D team simulated the force and power consumption of each joint, and designed 6 motors that simulate muscles. 3) Functionality: Simple object handling is possible, and Tesla expects the robot to learn to walk within a few weeks. In the future, Optimus will perform highly repetitive and dangerous work in factories. Through cost control, Tesla expects that users will be able to purchase robot products within 3-5 years, and the long-term price will be less than US$20,000. Although the overall progress has not exceeded market expectations, it is remarkable to complete the progress from concept to prototype in just one year.
　　FSD: The number of users of the beta version is growing rapidly, and global launch should be attainable by the end of the year. Compared to the 2021 version, the FSD Beta release is a huge step forward. 1) Application scale: The number of users using the FSD Beta software for autonomous driving has increased from about 2,000 in 2021 to 160,000 in Sep 2022, more than 75,000 neural network models have been trained, 4.8mn pieces of data have been accumulated, and 35 version updates. Elon Musk expects to roll out the beta version to the world by the end of the year. 2) Technical realization: Tesla collects a large amount of data and generates corresponding images, and processes the data after automatic or manual labeling, so as to continuously train the neural network model. At present, the Company's data set sources are divided into real data return and simulation data. The simulation data come from the simulation system and are used for model training for extreme cases. According to Tesla, the speed of generating virtual scenes is 1,000 times faster than before, and it can also add real elements such as signal lights and stop signs to be as close to the real world as possible. 3) Implemented functions: Based on a lot of training, the FSD Beta version can adapt to different road conditions in different countries. By predicting the actions of vehicles and pedestrians in the surrounding environment, autonomous driving can be achieved to a certain extent (such as automatic parking, moving between parking lots, waiting for traffic lights, etc.).
　　Supercomputer Dojo: Focusing on AI training and planning to launch the ExaPOD supercomputer in 1Q23. At present, Tesla has 30PB of video materials, and the large-scale supercomputer used for model training is mainly based on GPU. Compared with it, Dojo is different in the following respects: 1) Design idea: To match large-scale neural network training of autonomous driving, all hardware maintains high transmission bandwidth from the chip to the unit to the computer room, thereby reducing delay and performance loss. 2) Technical parameters: ①ExaPOD cluster: BF16/CFP8 peak computing power is 1.1EFLOPS, with 1.3T high-speed memory and 13TB high-bandwidth memory. ②System Tray parameters: 75mm height, 54 PFLOPS (BF16/CFP8), 13.4TB/S (half bandwidth), 100+KW Power. ③Standard interface processor parameters: 32GB (high bandwidth dynamic random access memory), 900TB/S (TTP bandwidth), 50GB/S (Ethernet bandwidth), 32GB/S (4th generation PCIE bandwidth). ④High-performance interface processor parameters: 640GB (high-bandwidth dynamic random access memory), 1TB/S (Ethernet bandwidth), 18TB/S (aggregate bandwidth to tiles). 3) Implemented functions: Tesla technical engineers predict that the Dojo supercomputing can increase the speed of network training by 30%. One Dojo training tile can replace 6 GPU Boxes, and the cost is lower than one GPU Box. The training time has been reduced from more than a month to within a week. The Company expects that the Dojo cabinet ExaPOD will be launched in the first quarter of 2023, and it may be provided to the market as a cloud service for model training in the future.
　　Potential risks: Macroeconomic growth comes under pressure; risk of loss of core technical personnel; AI product development progress is slower than expected; increased labor costs; intensified competition in the industry.
　　Investment advice: Tesla's stock price plunged recently due to a number of setbacks including the Fed’s continued tightening expectations, disappointing 3Q vehicle deliveries, little surprises from the AI Day, and Elon Musk’s restart of the Twitter acquisition. However, we consider the above setbacks more of a temporary disruption. In the medium to long term, Tesla's combined advantages of “automobile + software” could continue to support its rapid and highly-visible growth. We maintain our 2022E/23E/24E operating revenue forecasts of US$84.5bn/128.9bn/162.9bn, respectively, and non-GAAP net profit forecasts of US$12.13bn/20.15bn/27.96bn, respectively. The current price corresponds to 58x/35x/25x PE (non-GAAP). We remain optimistic about Tesla's short-, medium- and long-term investment value, and suggest investors actively tracking entry opportunities after the recent stock price correction.
【免责声明】本文仅代表第三方观点，不代表和讯网立场。投资者据此操作，风险请自担。

   【免责声明】本文仅代表第三方观点，不代表和讯网立场。投资者据此操作，风险请自担。