Elon Musk never envisioned Tesla as merely an automaker. Instead, he positioned the company as an artificial intelligence powerhouse focused on solving autonomous driving. Central to this ambitious vision was Tesla Dojo, a custom-built supercomputer designed specifically for training Full Self-Driving neural networks. This comprehensive timeline reveals the dramatic six-year journey of Tesla’s most ambitious computing project.
Tesla Dojo’s Initial Announcement and Early Development
Tesla first revealed Dojo during its 2019 Autonomy Day event. Musk introduced the supercomputer concept as a revolutionary approach to processing massive amounts of video data. The system promised to accelerate FSD development through specialized hardware. Importantly, Tesla announced that all current vehicles already contained the necessary hardware for full autonomy.
Building the Tesla Dojo Infrastructure
Tesla made significant progress throughout 2021-2022. The company unveiled its custom D1 chip and began assembling the first Dojo cabinets. Each tile contained 25 D1 chips, with production reaching one tile per day. Tesla demonstrated Dojo’s capabilities by running Stable Diffusion models live during AI Day presentations. The company projected completing seven Exapod clusters by early 2023.
Key Tesla Dojo Milestones
2019: Initial announcement at Autonomy Day
2021: D1 chip reveal and first AI Day demonstration
2022: First cabinet installation and load testing
2023: Production begins with $1 billion investment commitment
Tesla Dojo Faces Technical Challenges
Despite early progress, Tesla encountered significant obstacles. The company struggled with Nvidia GPU shortages and competing hardware priorities. Musk acknowledged Dojo as a “long-shot bet” during earnings calls. Tesla pursued a dual-path approach using both Nvidia hardware and Dojo simultaneously. This strategy revealed underlying uncertainties about Dojo’s capabilities.
The Rise of Cortex and Tesla Dojo’s Decline
By 2024, Tesla began shifting focus to Cortex, a new supercluster using Nvidia H100 GPUs. The company completed deployment of approximately 50,000 H100 units in Austin. Cortex enabled FSD V13 with major improvements in safety and comfort. Meanwhile, Tesla continued developing Dojo 2 while acknowledging convergence challenges between training and inference chips.
Tesla Dojo’s Final Chapter
In August 2025, Bloomberg reported Tesla had disbanded the Dojo team and shut down the project. Lead engineer Peter Bannon departed the company. Musk confirmed the shutdown on X, explaining that resource division between two chip designs became impractical. The AI6 chip emerged as Tesla’s unified solution for both inference and training needs.
Why Tesla Dojo Ultimately Failed
Resource allocation: Maintaining two chip designs proved too costly
Technical convergence: AI6 offered better integration possibilities
Market realities: Nvidia’s established ecosystem provided reliability
Strategic pivot: Focus shifted to unified architecture solutions
FAQs About Tesla Dojo
What was Tesla Dojo’s primary purpose?
Tesla Dojo was designed as a custom supercomputer to train the company’s Full Self-Driving neural networks using massive video data from Tesla vehicles.
When did Tesla first announce Dojo?
Elon Musk first teased Dojo during Tesla’s Autonomy Day event on April 22, 2019, describing it as a supercomputer for AI training.
Why did Tesla abandon the Dojo project?
Tesla discontinued Dojo to avoid dividing resources between two different AI chip designs, opting instead to focus on the AI6 chip that could handle both inference and training.
What replaced Tesla Dojo?
Tesla shifted to Cortex, a supercluster using Nvidia H100 GPUs, and is developing the AI6 chip as a unified solution for future AI training needs.
How much did Tesla invest in Dojo?
Tesla planned to spend over $1 billion on Dojo through 2024, with accumulated AI-related capital expenditures reaching approximately $5 billion.
Did Dojo ever become operational?
Yes, Dojo became operational and was running training tasks, but it never reached the scale originally envisioned by Tesla leadership.
