Tesla’s Dojo, a timeline

Elon Musk envisions Tesla as more than just an automotive manufacturer; he aspires for it to be a leader in artificial intelligence, particularly in developing self-driving technology. Central to this vision was the creation of Dojo, a powerful supercomputer engineered by Tesla to enhance its Full Self-Driving (FSD) neural networks. While FSD can automate several driving functions, it still necessitates vigilant human oversight. Tesla believes that with increased data, computing power, and training, it can bridge the gap from semi-autonomous to fully autonomous driving, with Dojo playing a pivotal role in this evolution. Musk has been promoting Dojo for years, intensifying discussions surrounding its capabilities throughout 2024. However, as Dojo's story unfolds, a new supercomputer named Cortex has emerged, shifting the focus within Tesla. Here’s a look back at the timeline of Dojo’s promises and developments. On April 22, during Tesla's Autonomy Day, the company showcased its AI team and discussed the technologies behind Autopilot and FSD. Musk unveiled Dojo as a supercomputer aimed at training AI, asserting that the hardware in all new Tesla vehicles was ready for full self-driving, pending only software updates. By February 2, Musk announced that Tesla would soon have over a million connected vehicles equipped with the necessary sensors for full autonomy, highlighting Dojo’s vast processing capabilities for video training data. He described it as a powerful tool poised to revolutionize FSD training. In August, Musk referred to Dojo as a 'beast,' indicating that the initial version would be ready approximately a year later, around August 2021. However, by December 31, he remarked that while Dojo was not essential, it would nonetheless significantly enhance self-driving capabilities. The official announcement of Dojo took place on August 19 during Tesla's inaugural AI Day, designed to attract talent to its AI team. The event also featured the introduction of the D1 chip, which was crucial for powering the Dojo supercomputer, with plans for a cluster containing 3,000 D1 chips. On October 12, Tesla published a whitepaper detailing Dojo Technology, focusing on a new type of binary floating-point arithmetic for deep learning applications. As time progressed, Musk indicated that Dojo would gradually phase in, reducing reliance on additional GPUs. Fast forward to June 21, where Musk confirmed that Dojo was operational and handling tasks in Tesla’s data centers. The company projected its computing capacity would rank among the world’s top five by February 2024, aiming for an impressive 100 exaflops by October 2024. However, by early 2025, the narrative began to shift dramatically. During a earnings call, Musk acknowledged the challenges surrounding Dojo, characterizing it as a high-risk, high-reward initiative. Plans for continued investment in the supercomputer were discussed, but there were also indications that Tesla was exploring alternatives. By August 7, reports surfaced that Tesla had disbanded the Dojo team, signaling a significant pivot away from the project. Musk confirmed the decision on social media, stating that with the convergence of resources towards the AI6 chip design, it was no longer viable to maintain Dojo as a separate entity. He remarked that all efforts would focus on the AI6 systems going forward. As Tesla's AI ambitions continue, the fate of Dojo serves as a reminder of the challenges and unpredictability inherent in the rapidly evolving field of artificial intelligence. The story of Dojo reflects both the aspirations and realities of Tesla's quest to redefine the future of mobility.