Skip to content

Tesla Abandons Dojo Supercomputer Development

Tesla has officially ended its Dojo supercomputer project, as confirmed by CEO Elon Musk over the weekend. Announced during the company's Autonomy Investor Day on April 22, 2019, Dojo, built using Tesla's custom D1 chip, was developed to train AI models for Full Self-Driving and various other...

Tesla Abandons the Development of the Dojo Supercomputer
Tesla Abandons the Development of the Dojo Supercomputer

Tesla Abandons Dojo Supercomputer Development

Tesla, the electric vehicle and clean energy company led by Elon Musk, has announced a significant shift in its AI hardware strategy. The original Dojo supercomputer project, designed to train Tesla's AI models for Full Self-Driving and other applications at a massive scale, has been officially discontinued [2][4]. However, the spirit and technological vision of Dojo live on through Tesla’s strategic pivot to the AI6 chip initiative.

The AI6 chip, a custom-designed unified AI processor, is intended to consolidate Tesla’s hardware needs for Full Self-Driving, the Optimus humanoid robot, and the AI training platform that Dojo originally aimed to realize [1]. AI6 represents a convergence in Tesla’s chip architecture — a “one chip for all” approach that replaces the earlier separated chip and supercomputer strategy with a scalable, modular solution [1][5].

According to Elon Musk and independent analysis, AI6 will be produced by Samsung in a large-scale $16.5 billion fabrication deal, with manufacturing in Texas [1]. AI6 is designed for both neural network training and inference, emphasizing high memory bandwidth, extensibility, and compatibility with mainstream ML frameworks [4].

Tesla envisions building its next-generation AI training supercomputer (informally “Dojo 3”) by clustering AI6 modules alongside GPUs from partners such as Nvidia and AMD — a hybrid architecture combining Tesla’s custom silicon with proven external hardware to achieve state-of-the-art training performance without the complexity of the original Dojo [4].

Elon Musk described AI6’s performance as a significant step forward beyond prior chips, such as the AI4 and AI5, and confirmed the AI6 project as the functional successor to Dojo [3][5]. The AI6 system-on-chips may arguably live on in the form of large boards, following the ultra-fast interconnect design of Dojo's unique system-on-wafer.

The former Dojo lead, Ganesh Venkataramanan, left the project less than two years before its scrapping. The reasons for the project's termination have not been provided. Tesla announced plans for a $500 million Dojo-powered supercomputer in Buffalo, New York, in January 2024 [6].

In summary, while the original Dojo supercomputer project has been disbanded, its spirit and technological vision have shifted into the AI6 chip initiative. AI6 embodies the next-generation AI hardware Tesla will deploy across its product ecosystem and for AI training, representing a strategic shift from bespoke supercomputer construction to a modular, scalable chip-centric platform optimized for both in-car AI and data center training workloads [2][3][4][5].

Read also:

Latest