Tesla Abandons In-House Chip Development for Driverless Technology, Shifts Focus to Nvidia and External Partners

Tesla is disbanding the team responsible for its Dojo supercomputer project, marking an end to the automaker’s endeavors in developing in-house chips for autonomous technology, according to reports from Bloomberg. Peter Bannon, the lead of the Dojo team, is departing the company, with remaining members being reassigned to other data center and compute projects within Tesla.

The dissolution of Tesla’s Dojo efforts coincides with the exit of approximately 20 employees who have founded a new AI startup named DensityAI. The venture, expected to emerge from stealth soon, is developing chips, hardware, and software designed for data centers powering AI in robotics, AI agents, and automotive applications. Ganesh Venkataramanan, former head of Dojo, along with ex-Tesla employees Bill Chang and Ben Floering, established DensityAI.

This move occurs at a critical juncture for Tesla as CEO Elon Musk pushes for shareholders to view the company as an AI and robotics powerhouse. Despite a limited launch of robotaxis in Austin last June featuring Model Y vehicles manned by humans, which resulted in several reports of problematic driving behavior, Musk continues to advocate for Tesla’s transformation into an autonomous driving leader.

Tesla’s decision to shutter Dojo, a project Musk has discussed since 2019, represents a significant shift in strategy. He had previously stated that Dojo would serve as the foundation of Tesla’s AI aspirations and its goal to achieve full self-driving due to its capacity to process immense volumes of video data.

Morgan Stanley predicted in 2023 that Dojo could potentially add $500 billion to Tesla’s market value by tapping into new revenue streams such as robotaxis and software services. Musk emphasized the importance of Dojo ahead of Tesla’s robotaxi unveiling in October last year, stating that the company’s AI team would intensify efforts on Dojo.

However, discussions surrounding Dojo ceased around August 2024 when Musk began promoting Cortex instead, a “giant new AI training supercluster” being built at Tesla’s headquarters in Austin. The project aimed to address real-world AI challenges.

Dojo encompassed both supercomputer and in-house chip development. In 2021, Tesla unveiled its D1 chip as part of the Dojo announcement. Venkataramanan showcased the chip, which Tesla claimed would work alongside Nvidia’s GPU to power the Dojo supercomputer. The automaker also mentioned plans for a next-gen D2 chip intended to address any information flow bottlenecks of its predecessor.

Sources told Bloomberg that Tesla is now planning to increase its dependence on external partners like Nvidia, AMD for compute, and Samsung for chip manufacturing. Last month, Tesla signed a $16.5 billion deal with Samsung for the production of AI6 inference chips, a design promising scalability from powering full self-driving (FSD) and Tesla’s Optimus humanoid robots to high-performance AI training in data centers.

During Tesla’s second-quarter earnings call, Musk hinted at potential overlaps between Dojo 3 and the AI6 inference chip.

The news comes as Tesla’s board offers Musk a $29 billion compensation package to remain with the company and further advance its AI initiatives, rather than being distracted by his other ventures such as xAI, a pure-play AI startup.