Tesla shuts Dojo supercomputer project; 20 engineers join former head Ganesh Venkataramanan

Tesla has shut down its Dojo supercomputer team and ended in-house chip development for autonomous driving, according to a Bloomberg report citing people familiar with the matter. The decision includes the departure of Dojo leader Peter Bannon, with remaining team members reassigned to other data centre and compute projects.

The move follows the exit of around 20 engineers from the Dojo team, who have joined DensityAI, a startup founded by former Dojo head Ganesh Venkataramanan along with ex-Tesla employees Bill Chang and Ben Floering. DensityAI is preparing to exit stealth mode, focusing on chips, hardware, and software for AI-driven data centres in robotics, AI agents, and automotive applications. The company plans to develop scalable, energy-efficient high-performance computing solutions for next-generation AI workloads.

Dojo, first announced in 2019, was intended to support Tesla’s AI ambitions, particularly in full self-driving (FSD), by processing large volumes of video data. It featured the in-house D1 chip, unveiled in 2021, which was designed to work alongside Nvidia GPUs. A second-generation D2 chip was planned to address data bottlenecks. In 2023, Morgan Stanley estimated that Dojo could add $500 billion to Tesla’s market value through robotaxi and software revenue.

Tesla’s decision to wind down Dojo comes as CEO Elon Musk shifts focus towards Cortex, a new AI training supercluster at Tesla’s Austin headquarters. This follows criticism of Tesla’s limited robotaxi launch in June 2025, which involved Model Y vehicles with human co-pilots and reports of erratic driving.

Tesla is now leaning more on external partners. Last month, it signed a $16.5 billion deal with Samsung to produce AI6 inference chips for applications ranging from FSD and Optimus humanoid robots to AI training systems. The company is also expanding partnerships with Nvidia for compute resources and AMD for chip solutions. Musk has indicated plans to streamline future AI hardware projects to reduce overlap.