## Tesla’s Dojo: A Timeline of AI Ambition
Tesla’s Dojo supercomputer, a monumental undertaking in artificial intelligence, is designed to rapidly train neural networks for autonomous driving and beyond. Its development has been a journey marked by bold announcements and significant engineering challenges.
**2019:** Elon Musk first hints at Tesla developing its own custom AI training hardware, recognizing the limitations of off-the-shelf GPUs for their specific needs.
**2020:** The name “Dojo” emerges, referring to an in-house developed supercomputer tailored for training the massive datasets required for Full Self-Driving (FSD).
**August 2021 (AI Day 2021):** Tesla officially unveils the D1 chip, the custom processor that forms the heart of Dojo. They detail its “tile” architecture, designed for high bandwidth and low latency, and outline the vision for interconnecting these into a powerful, scalable supercomputer.
**Early 2022:** Tesla begins the physical build-out of Dojo clusters, with initial modules and cabinets being assembled at their data centers.
**October 2022 (Q3 Earnings):** Elon Musk confirms that the first Dojo clusters are starting to come online, signaling a transition from design to operational testing.
**July 2023 (Q2 Earnings):** Tesla provides a significant update, stating Dojo is “training” and is projected to require over $1 billion in investment by the end of 2024. The focus remains on internal development to accelerate FSD.
**Late 2023 – Present:** Tesla continues to scale Dojo, integrating more D1 chips into larger “ExaPODs” to achieve exaFLOP-level performance. The system is incrementally brought online, and actively utilized to refine and advance Tesla’s AI models for vehicle autonomy.
Dojo represents Tesla’s commitment to vertical integration in AI, aiming to outpace competitors by designing specialized hardware optimized for their unique machine learning workloads.