Tesla Dojo: Elon Musk’s big plan to build an AI supercomputer, explained
Elon Musk is developing the Dojo supercomputer to enhance Tesla's Full Self-Driving (FSD) capabilities and potentially boost Tesla's revenue by $500B.
Elon Musk is focusing on building the Dojo supercomputer to propel Tesla's AI ambitions, especially around Full Self-Driving (FSD). The Dojo system is designed to train the FSD neural networks using video data collected from around 2 million Tesla vehicles currently equipped with the FSD feature, which still requires human intervention for now. Musk envisions Dojo, powered by the company's custom D1 chips, to become more efficient than using Nvidia GPUs, which Tesla heavily relies on at the moment.
The proprietary D1 chips, manufactured by Taiwan Semiconductor Manufacturing Company (TSMC), promise higher bandwidth and lower latencies, aiming to optimize AI workloads. Each Dojo tile, composed of 25 D1 chips, achieves 9 petaflops of compute power and 36 terabytes per second of bandwidth. Tesla plans to scale Dojo through multiple ExaPODs and is already working on a next-gen D2 chip to improve information flow by placing it onto a single silicon wafer.
In addition to boosting Tesla's self-driving technology, the Dojo supercomputer has the potential to create new business opportunities for the company. By taking control of its own chip production, Tesla aims to reduce dependency on Nvidia's increasingly costly hardware. A Morgan Stanley report predicts that Dojo could add $500 billion to Tesla's valuation by unlocking new revenue streams such as robotaxis and AI software services.