## Tesla’s Dojo: A Timeline of Innovation
Tesla’s Dojo, the custom-built supercomputer designed to train its vast neural networks for autonomous driving, represents a significant leap in AI infrastructure. Here’s a brief timeline of its development:
**August 2020:** Elon Musk first hints at “Project Dojo” during Tesla’s AI Day, describing it as a supercomputer to process video data for self-driving.
**August 2021:** At AI Day 2021, Tesla officially unveils detailed plans for Dojo. They introduce the custom-designed D1 chip, the “Training Tile” architecture (composed of multiple D1 chips), and the overall system design aiming for unprecedented compute density and bandwidth.
**Late 2021/Early 2022:** Initial production of the D1 chips by TSMC is reported, with Tesla beginning to receive and test the first silicon.
**October 2022:** Elon Musk states that Dojo is expected to become operational and begin internal training by the end of the year, though scaling would be an ongoing process.
**July 2023:** Reports emerge that Dojo is starting to come online and is actively being used by Tesla internally for FSD model training, albeit not yet at its full planned scale.
**Q3 2023:** Tesla emphasizes Dojo’s growing importance, with Musk highlighting its potential as a significant competitive advantage as it continues to be built out and optimized.
**Ongoing:** Dojo’s development continues, with Tesla continually expanding its capacity and integrating it more deeply into the FSD software stack, aiming to accelerate the path to full self-driving capabilities.