Tesla‘s ambitious AI project, Dojo, designed to revolutionize the company’s Full Self-Driving capabilities, has experienced a dramatic rise and fall. Elon Musk initially touted Dojo as a crucial piece in Tesla’s path to producing fully autonomous vehicles and humanoid robots. This heavily-customized supercomputer was built especially for the purpose of training neural networks. These foundational shifts in strategy led to Tesla shuttering their AI division by August 2025. This decision represented a significant change in the firm’s longstanding resistance to artificial intelligence.
Dojo’s journey began with a strong vision from Musk, who outlined the potential of the supercomputer at Tesla’s 2021 AI Day. Ganesh Venkataramanan, senior director of Autopilot hardware, demonstrated the D1 training tile. He had a lot of exciting things to say about the innovative depths of Dojo. Dojo was built to supercharge Tesla’s robotaxi and full self-driving dreams. It was supposed to usher the company—and the world—into a new era of autonomous driving.
Musk’s confidence in Dojo waxed and waned over the years. He had first called into question the feasibility of the complicated project, then, after its defeat, citing it an “evolutionary dead end.” That declaration the following year presaged the disbanding of the team behind Dojo and the shutting down of the company.
The Vision Behind Dojo
Tesla originally imagined Dojo to be a bedrock of its AI ambitions. Tesla developed the supercomputer to power its Full Self-Driving neural networks. It required staggering amounts of data and processing power to accomplish it. Musk promised that full autonomy and the rollout of robotaxis depended on the success of this project.
In June 2023, Musk declared that Dojo had been operational for the past several months. In doing so, it tackled a number of important, useful tasks. The supercomputer was key to training the algorithms. These algorithms allow even Tesla’s most basic vehicles to autonomously navigate suburban streets and freeway interchanges without human intervention.
Dojo’s D1 chip is outfitted with a whopping 50 billion transistors. This die is a massive 645 millimeters squared and is an example of the best-in-class technology that could soon power Tesla’s AI dreams. As Tesla attempted to secure its position in the competitive AI landscape, Musk announced plans for a second supercluster, known as Dojo 2, utilizing in-house D2 chips.
“Thinking about Dojo 3 and the AI6 inference chip, it seems like intuitively, we want to try to find convergence there, where it’s basically the same chip,” – Elon Musk
As he did so, Musk’s optimism turned to reality as he started recognizing the company’s priorities began to change.
The Shift in Strategy
Just this past August 2024, Tesla made a historic deal with Samsung. The $16.5 billion deal secured next-generation AI6 chips as recently as a few weeks before they spun out Dojo. This move indicated a strategic pivot towards newer technology that would reportedly streamline operations and enhance AI capabilities.
In the months that followed, Musk himself explained the justification for the sudden shuttering of Dojo. He explained that every route in the situation has led to AI6 technology. This challenge required him to make some painful personnel decisions. The rest of the team took the courageous step to stop further development on Dojo. This was a smart move to refocus scarce resources on more commercially viable prospects.
“Once it became clear that all paths converged to AI6, I had to shut down Dojo and make some tough personnel choices, as Dojo 2 was now an evolutionary dead end,” – Elon Musk
Along with this big strategic pivot, Tesla kicked off the announcements about its new AI training supercluster, Cortex, publicly in August of 2024. Cortex had the ambitious goal of making video training more accessible, not just for Full Self-Driving technology but for the exciting new Optimus humanoid robot project.
The Future of Tesla’s AI Endeavors
Even though Dojo is now closed, continued on potentially changing the world through Omega AI.Cortex at Gigafactory Texas, now finished, displaying just under 50,000 H100 training units. This accomplishment marks a significant step forward for the company. This new cluster is believed to have played an essential role in achieving version 13 of Tesla’s supervised Full Self-Driving software.
Morgan Stanley recently released a report projecting that advancements in Tesla’s AI capabilities could potentially add $500 billion to the company’s market value by unlocking new revenue streams. Forecasts like these only highlight just how crucial AI training systems are to keeping Tesla’s leg up over its automotive competitors.
Aside from the $1.5 billion investment in OpenAI, Musk has been bullish on what is possible with Tesla AI technology. He made clear that this would be directed to shoring up their own AI talent. This buildup will lead to the long-expected unveiling of the robotaxi in October 2024.
“A path to being competitive with Nvidia with Dojo,” – Elon Musk
This hopeful projection is undoubtedly clouded by the challenges of data availability and model training effectiveness. Anand Raghunathan, an expert in the AI training pipeline practices, emphasized possible economic limitations. He stressed the need for high-quality data to improve model performance.
“Some people claim that we might actually run out of meaningful data to train the models on. More data doesn’t necessarily mean more information,” – Anand Raghunathan
