Tesla Disbands Dojo AI Supercomputer Team, Marking Strategic Shift in Self-Driving Efforts

Tesla is dismantling its Dojo AI training supercomputer project, marking a significant retreat from its earlier ambition to build a fully in-house AI infrastructure for full self-driving technology. According to Bloomberg, citing anonymous sources, the team behind Dojo is being disbanded, with its lead, Peter Bannon, departing the company. Remaining team members are being reassigned to other data center and compute initiatives within Tesla. The move follows the departure of around 20 employees who left Tesla to co-found DensityAI, a new AI startup reportedly emerging from stealth. The company is led by Ganesh Venkataramanan, former head of Dojo, along with ex-Tesla engineers Bill Chang and Ben Floering. DensityAI is developing custom chips, hardware, and software aimed at powering AI data centers for robotics, AI agents, and automotive applications. This shift comes at a pivotal moment for Tesla. Despite CEO Elon Musk’s long-standing vision of positioning the company as an AI and robotics leader, Tesla’s progress has stalled. The limited robotaxi rollout in Austin last June, which required a human safety driver in the passenger seat, was marred by incidents of erratic vehicle behavior, undermining confidence in the technology. Dojo, first announced in 2021 during Tesla’s inaugural AI Day, was meant to be the backbone of Tesla’s AI ambitions. Musk touted it as essential for processing massive volumes of video data to achieve full self-driving. The project combined custom-built D1 chips with a specialized supercomputer architecture. Tesla had also been developing a next-generation D2 chip to overcome data flow limitations. However, momentum for Dojo faded around August 2024, when Musk began promoting a new initiative called Cortex—a massive new AI training cluster being built at Tesla’s Austin headquarters. This pivot signals a strategic retreat from building proprietary hardware and a renewed reliance on external partners. Tesla is now increasing its use of Nvidia GPUs and expanding partnerships with AMD for compute and Samsung for chip manufacturing. In a major development, Tesla recently signed a $16.5 billion deal with Samsung to produce its AI6 inference chips. These chips are designed to scale across applications, from Tesla’s Full Self-Driving system and Optimus humanoid robots to high-performance AI training in data centers. During Tesla’s second-quarter earnings call, Musk hinted at internal restructuring, suggesting that future efforts around Dojo 3 and the AI6 chip could converge on a single, unified chip design. This indicates a shift toward leveraging existing industry technology rather than building from the ground up. The decision comes as Tesla’s board prepares to offer Musk a $29 billion compensation package aimed at keeping him focused on Tesla’s AI and robotics goals, rather than diverting attention to his other ventures like xAI. The move underscores the growing pressure on Tesla to deliver on its AI promises amid intense competition from OpenAI, Google, and other AI-focused companies. TechCrunch has reached out to Tesla for comment.