Tesla Shuts Down Dojo AI Supercomputer - Full Self-Driving Impact

Tesla Discontinues In-House Chip Development for Autonomous Driving

According to a report from Bloomberg, Tesla is dismantling the team responsible for its Dojo supercomputer, effectively ending the company’s initiative to create its own chips for self-driving technology.

Leadership Transition and Team Reassignment

Peter Bannon, the leader of the Dojo project, is departing from Tesla. The remaining personnel will be integrated into other data center and computational projects within the organization, as indicated by sources who requested anonymity.

Departure of Key Personnel and Formation of DensityAI

This decision follows the recent exit of approximately 20 employees who have established a new AI company, DensityAI. This startup is preparing to publicly launch and is focused on developing chips, hardware, and software solutions for AI-powered data centers utilized in robotics, AI agents, and automotive applications.

Timing and Strategic Implications

The restructuring of Tesla’s Dojo efforts occurs during a critical period for the company.

Shift in Focus Towards AI and Robotics

CEO Elon Musk has been actively encouraging investors to perceive Tesla as an AI and robotics firm. However, the initial rollout of a robotaxi service in Austin, Texas, in June experienced issues, with reports of problematic driving behavior from vehicles that still required a human safety driver.

A Change in Long-Term Strategy

Tesla’s choice to discontinue Dojo represents a significant strategic alteration. Musk had previously emphasized that Dojo would be central to Tesla’s AI aspirations and the achievement of full self-driving capabilities, citing its capacity to process extensive video data.

Previous Projections and Expectations

In 2023, Morgan Stanley estimated that Dojo could potentially increase Tesla’s market value by $500 billion by enabling new revenue streams from robotaxis and software services. Musk had also stated that the AI team would intensify its focus on Dojo prior to the unveiling of the robotaxi, which took place in October.

Transition to Cortex

Discussions surrounding Dojo diminished around August 2024, as Musk began promoting Cortex, described as Tesla’s “giant new AI training supercluster” being constructed at the company’s headquarters in Austin to address real-world AI challenges.

The D1 and D2 Chips

The Dojo project encompassed both supercomputing infrastructure and in-house chip development. Tesla introduced its D1 chip during the inaugural AI Day in 2021. Ganesh Venkataramanan presented the chip, which was intended to work in conjunction with Nvidia GPUs to power the Dojo supercomputer. Development of a next-generation D2 chip was also underway, aimed at resolving data flow limitations of the initial design.

Increased Reliance on External Partners

Sources indicate that Tesla now intends to increase its dependence on Nvidia, as well as other technology partners such as AMD for computing and Samsung for chip manufacturing. A $16.5 billion agreement was recently signed with Samsung to produce Tesla’s AI6 inference chips, designed to support a range of applications from FSD and the Optimus humanoid robot to high-performance AI training.

Potential Redundancies and Chip Convergence

During Tesla’s second-quarter earnings call, Musk alluded to potential overlaps in functionality. He suggested a desire to converge the designs of Dojo 3 and the AI6 inference chip into a unified solution.

Musk’s Compensation and AI Focus

This news coincides with Tesla’s board approving a $29 billion compensation package for Musk, intended to retain his leadership and drive the company’s AI initiatives forward, while minimizing distractions from his other ventures, including xAI.

TechCrunch has contacted Tesla for further comment.

Have a sensitive tip or confidential documents? We’re reporting on the inner workings of the AI industry — from the companies shaping its future to the people impacted by their decisions. Reach out to Rebecca Bellan at rebecca.bellan@techcrunch.com and Maxwell Zeff at maxwell.zeff@techcrunch.com. For secure communication, you can contact us via Signal at @rebeccabellan.491 and @mzeff.88.