Tesla's Dojo: A Journey Towards Full Autonomy

Elon Musk envisions Tesla as more than just an automaker; he aims to transform it into an AI powerhouse capable of creating fully autonomous vehicles. Central to this vision is Dojo, Tesla's custom-built supercomputer designed to train its Full Self-Driving (FSD) neural networks. While FSD isn't yet fully autonomous and still requires human oversight, Tesla believes that with more data, computational power, and training, it can achieve true self-driving capabilities. Here's a timeline of Dojo's development and milestones.

2019: The Beginning

•  April 22: During Tesla's Autonomy Day, the AI team discusses Autopilot and FSD, highlighting Tesla's custom-built chips for neural networks and self-driving cars. Musk introduces Dojo as a supercomputer for AI training, stating that all Tesla cars produced at the time had the necessary hardware for full self-driving, pending a software update.

2020: Building Momentum

•  February 2: Musk announces that Tesla will soon have over a million connected vehicles with the sensors and computing power needed for full self-driving, emphasizing Dojo's capabilities.

•  August 14: Musk reiterates the plan to develop Dojo to process vast amounts of video data, calling it "a beast" and projecting its launch around August 2021.

•  December 31: Musk states that while Dojo isn't essential, it will significantly enhance self-driving capabilities, aiming for Autopilot to be more than ten times safer than human drivers.

2021: Official Launch

•  August 19: Tesla officially announces Dojo at its first AI Day, introducing the D1 chip to power the supercomputer. The AI cluster will house 3,000 D1 chips.

•  October 12: Tesla releases a Dojo Technology whitepaper outlining a new type of binary floating-point arithmetic for deep learning neural networks.

2022: Progress and Testing

•  August 12: Musk announces plans to phase in Dojo, reducing the need for incremental GPUs.

•  September 30: At Tesla's second AI Day, the first Dojo cabinet is installed, and the company demonstrates Dojo running a Stable Diffusion model to create an AI-generated image of a "Cybertruck on Mars." Tesla aims to complete a full Exapod cluster by Q1 2023 and plans to build seven Exapods in Palo Alto.

2023: A High-Stakes Bet

•  April 19: Musk tells investors that Dojo could significantly reduce training costs and become a sellable service, comparing it to Amazon Web Services.

•  June 21: Tesla's neural networks are already in customer vehicles, with Dojo projected to be among the top five compute systems globally by February 2024.

•  July 19: Tesla starts Dojo production and plans to spend over $1 billion on it through 2024.

•  September 6: Musk highlights the challenges of managing data from Tesla's vehicles and the role of Nvidia and Dojo in addressing these challenges.

2024: Scaling Up

•  January 24: Musk acknowledges Dojo as a high-risk, high-reward project, with plans for multiple versions of Dojo.

•  January 26: Tesla announces a $500 million investment to build a Dojo supercomputer in Buffalo.

•  April 30: TSMC reveals that Dojo's next-generation training tile, the D2, is already in production.

•  May 20: Musk notes that the Giga Texas factory extension will include a super dense, water-cooled supercomputer cluster.

•  June 4: A CNBC report reveals Musk diverted Nvidia chips reserved for Tesla to X and xAI, with plans to house 50,000 H100s for FSD training.

•  July 1: Musk indicates that current Tesla vehicles may need hardware upgrades for the next-gen AI model.

•  July 23: Musk discusses Nvidia supply challenges and the need to focus on Dojo to ensure sufficient training capability.

•  July 30: Musk mentions that AI5 is about 18 months away from high-volume production.

•  August 3: Musk tours the Tesla supercompute cluster at Giga Texas, which will include around 100,000 Nvidia GPUs.