Tesla's Dojo, a timeline

Core Viewpoint - Tesla aims to transition from being solely an automaker to an AI company, focusing on achieving full self-driving capabilities through advanced computing power and data processing [1][2]. Development of Dojo - Dojo was introduced as a custom-built supercomputer designed to train Tesla's Full Self-Driving (FSD) neural networks, which at the time required human oversight despite some automated capabilities [2][3]. - The timeline of Dojo's development includes its first mention in 2019, with Musk highlighting its potential to process vast amounts of video data for training AI [4][5][8]. - By 2021, Tesla officially announced Dojo, introducing its D1 chip and outlining plans for a supercomputer capable of significant AI training [9][10]. Progress and Challenges - Throughout 2022 and 2023, Tesla reported progress on Dojo, including the installation of its first cabinet and plans for a full Exapod cluster by early 2023 [10][12]. - Musk indicated that Dojo could significantly reduce training costs and potentially become a sellable service, similar to Amazon Web Services [11][12]. - However, by mid-2023, Tesla faced challenges with Nvidia hardware supply, prompting a renewed focus on Dojo to ensure adequate training capabilities [16]. Transition to Cortex - In 2024, Tesla began transitioning from Dojo to a new supercomputer called Cortex, which utilizes Nvidia GPUs and aims to enhance AI training for FSD [18][19]. - The Cortex supercomputer was reported to consist of approximately 50,000 H100 Nvidia GPUs, facilitating improvements in FSD performance [19][20]. - By early 2025, the Dojo project was officially shut down, with Tesla consolidating its resources towards the development of the AI6 chip, which is intended to serve multiple AI applications [22][23]. Future Directions - Tesla's future plans include scaling AI capabilities with the AI6 chip, which is designed for both inference and training, indicating a strategic shift in its AI development approach [22][23]. - The company aims to maintain a competitive edge in AI by focusing on integrated chip designs rather than dividing resources across different projects [23].