} ?>
BOC Securities released a research report saying that according to Tesla's computing power development plan released in June 23, Dojo will become the world's top five computing power facilities in the first quarter of next year and reach a super computing power of 100EFlops in October next year. In the future, Tesla will vigorously lay out the driverless taxi platform, using a service model similar to Uber or shared apartment company Aibiying, taking a corresponding share from each order, and the increase in the proportion of FSD software services is expected to significantly boost the company's gross profit level. Robots and self-driving cars are the same, Tesla humanoid robot Optimus equipped with FSDAI computing chip DojoD1, the future is expected to be linked with the self-driving business, further open up growth space.
The main points of view of BOC Securities are as follows:
Tesla Dojo provides AI business computing power support base.
according to the speech of horizon co-founder and CTO Huang Chang in "electric car observer", currently L2 level automatic driving requires about 10TOPS of computing power, L3 requires more than 100TOPS, and L4 requires 1000TOPS. Dojo uses a self-computing power chip to provide "unsupervised" labeling and training for autonomous driving using massive amounts of video data. According to the calculation power expectation chart released by Tesla's artificial intelligence Twitter account "TeslaAI", Tesla's calculation power scale will enter the top five in the world in February 2024, and the total scale will reach 100Exa-Flops in October, which is equivalent to the total calculation power of 300000 Avida A100 graphics cards. Dojo can improve Tesla's data analysis and labeling capabilities, thereby accelerating its automatic driving system iteration process.
vigorously layout driverless taxi platform, software services accounted for increased is expected to significantly boost the company's gross margin.
Recently, Tesla's head of investor relations Martin Veccha said at the Goldman Sachs Technology Summit that the Robotaxi (driverless taxi) business will become Tesla's third-generation platform. In addition to selling cars, Tesla will also take a corresponding share from each order in a similar Uber model. The driverless taxi platform business is expected to significantly improve the company's profit level. At present, Tesla's overall business gross margin is about 20%, while according to Tesla's 2022 annual report, the gross margin of deferred revenue realized by FSD is as high as 90%. The FSD software services business is expected to be the company's biggest attraction in the future.
FSD is expected to achieve fully autonomous driving by the end of the year.
Recently, Musk demonstrated the FSDBetaV12 system in a live broadcast. Through a large amount of video training data and Nvidia's H100GPU blessing, the system realized end-to-end AI automatic driving, and used neural networks to identify obstacles such as traffic lights, pedestrians and other vehicles, and the overall operation was relatively smooth. According to Musk's speech at Tesla's 2023 second quarter earnings communication meeting, the new FSD will be fully autonomous by the end of this year.
Humanoid robots are expected to open up further room for growth.
Musk stated at the Tesla Investor Day event held on March 1, 23 that self-driving cars are essentially the same as robots. Tesla's humanoid robot Optimus is equipped with FSD's AI chip DojoD1 and uses FSD-related vision, positioning and other systems. One of the reasons that currently restricts the large-scale commercialization of humanoid robots is their high price. FSD uses BEV + Transformer pure vision algorithm path, the robot also relies on computer vision for perception, and the synergy between automatic driving and humanoid robot is strong. With the large-scale reuse of robots and FSD algorithms, the Optimus cost is expected to be significantly reduced.
Investment advice:
it is suggested to pay attention to hongsoft technology (688088.SH)(Bev + Transformer vision algorithm), light court information (301221.SZ) (Huawei automatic driving data processing), jingwei hengrun (688326.SH) (covering the whole area automatic driving scene below L4), zhongke chuangda (300496.SZ) (automatic driving operating system), desai xiwei (002920.SZ)(automatic driving controller) -00020 (automatic driving model).
Risk Warning: Dojo technology breakthrough is less than expected, FSD landing process is less than expected, policy promotion is less than expected.
Ticker Name
Percentage Change
Inclusion Date