Tesla Discontinues Dojo Supercomputer Project, Redirects Staff According to Recent Reports
Tesla has announced the shutdown of its Dojo supercomputer project and the disbanding of the team working on it, as of August 2025. The decision comes after Elon Musk concluded that the Dojo 2 was an "evolutionary dead end" following the company's pivot to the AI6 chip architecture [1][2][5].
The Dojo project was initially designed to build an in-house supercomputer with proprietary chips (D1, D2, and planned D3 chips) to accelerate Full Self-Driving (FSD) algorithm training and reduce dependence on external chip suppliers [1]. However, technical and strategic challenges, as well as key personnel departures, led to the project's closure.
Tesla is now focusing its resources on the AI5 and AI6 chips, which are being manufactured externally by TSMC and Samsung. The AI5 chip targets Tesla's FSD system, while AI6 is designed for onboard inference in vehicles and humanoid robots, as well as large-scale AI training [1][3][5]. Musk explained that it made no sense to split resources between Dojo chips and the AI5/AI6 chip line, which are intended for both training and inference, and can be scaled more efficiently by putting multiple AI5/AI6 chips on a single board for supercomputing tasks [5].
This shift reflects a move towards a hybrid AI training architecture combining Tesla-designed AI6 chips and third-party GPUs like Nvidia's for scalable AI training infrastructure [4]. The decision follows the recent exit of about 20 Dojo engineers who launched a new AI venture, DensityAI, with the aim of developing chips and hardware solutions for data centers and AI applications in robotics and autonomous systems [6].
Peter Bannon, who led the Dojo initiative, is departing the company. The CEO of Tesla, Elon Musk, desires to align future hardware like the Dojo 3 and AI6 inference chip into a unified architecture [5]. A follow-up D2 version of the Dojo project was under development to improve performance and address bottlenecks [1].
Elon Musk has hinted at the possibility of consolidating AI chip development with partner technologies, such as the $16.5 billion semiconductor deal with Samsung Electronics to manufacture its AI6 chips through 2033 [7]. The AI6 chips, manufactured by Samsung, are expected to support various applications, including Tesla's FSD platform, Optimus humanoid robots, and large-scale AI training [7].
Remaining members of the Dojo team are being reassigned to other data center and computing projects within Tesla. At its peak, the Dojo project featured Tesla's custom D1 chip [1].
[1] Tesla's Dojo AI Chip Project Shut Down
[2] Tesla Disbands Dojo Supercomputer Team
[3] Tesla AI5 and AI6 Chips Explained
[4] Tesla's Dojo Supercomputer: What Happened?
[5] Elon Musk on the Future of Tesla's Hardware
[6] DensityAI: A New Venture by Ex-Tesla Dojo Engineers
[7] Tesla's $16.5 Billion Semiconductor Deal with Samsung
- Despite the shutdown of the Dojo supercomputer project, Tesla continues to prioritize finance and technology by focusing on the development of its AI5 and AI6 chips, especially the AI6 for onboard inference in vehicles and humanoid robots.
- The decision to shift towards a hybrid AI training architecture involving Tesla-designed AI6 chips and third-party GPUs like Nvidia's is one of Elon Musk's strategic moves in the realm of finance and technology, aligning with his vision for a unified hardware architecture in the future.
