Tesla Disbands Dojo AI Training Supercomputer Team, Shifts Focus to AI5 and AI6 Chips

Tesla has disbanded its Dojo AI training supercomputer team and shelved its D2 chip development after Elon Musk determined it was an "evolutionary dead end". The company will focus on its AI5 and AI6 chips, manufactured by TSMC and Samsung, respectively, which will power FSD and self-driving capabilities. Musk has also confirmed that a new supercomputer cluster will be built using many AI5/AI6 chips on a board, referred to as "Dojo 3".

In a significant strategic pivot, Tesla has disbanded its Dojo AI training supercomputer team and shelved the development of its D2 chip. The move, announced by CEO Elon Musk, signals a shift in focus towards AI5 and AI6 chips, which will power Tesla's Full Self-Driving (FSD) and autonomous driving capabilities. Musk has also confirmed that a new supercomputer cluster, referred to as "Dojo 3," will be built using many AI5/AI6 chips on a board.

The decision to disband the Dojo team comes after several years of development, during which the project aimed to create an in-house AI training infrastructure using Tesla's custom D1 chip. However, the project faced persistent delays and a talent drain, with key personnel, including Peter Bannon and Ganesh Venkataramanan, leaving to form DensityAI [1]. The exit of these engineers and the formation of DensityAI, a new AI startup, further underscores the challenges faced by the Dojo project.

Tesla's new focus on AI5 and AI6 chips represents a pragmatic pivot towards efficiency and cost savings. The AI5 chip, expected to enter production by late 2026, will be manufactured by TSMC, while the AI6 chip, scheduled for subsequent years, will be produced by Samsung. This shift allows Tesla to consolidate its AI development efforts, streamline R&D costs, and accelerate deployment timelines for FSD and Optimus.

The partnership with Samsung, valued at $16.5 billion, is particularly noteworthy. By diversifying chip sourcing beyond TSMC, Tesla mitigates supply chain risks and secures long-term production capacity through 2033. This move also aligns with Musk's broader vision of integrating xAI (his AI venture) with Tesla's robotics and autonomous systems, leveraging X's data for training.

While the disbanding of the Dojo team may disappoint those who viewed it as a potential $500 billion market cap catalyst, it also eliminates a costly, resource-intensive project that had outlived its strategic utility. The inference-first strategy, prioritizing real-time data processing, positions Tesla to outperform competitors in speed to market and cost efficiency.

References:
[1] https://driveteslacanada.ca/news/tesla-shuts-down-dojo-supercomputer-program-reassigns-team-report/
[2] https://timesofindia.indiatimes.com/technology/tech-news/tesla-shuts-down-its-ai-training-supercomputer-dojo-loses-20-key-execs-to-company-founded-by-ganesh-venkataramanan/articleshow/123223061.cms
[3] https://www.ainvest.com/news/tesla-ai-autonomous-vehicle-strategy-reassessment-light-restructuring-2508/