Tesla’s ambitions for its Dojo supercomputer have suddenly hit the brakes.
After years of fanfare around its in-house AI hardware, the electric vehicle company is disbanding the team responsible for Dojo and shifting focus elsewhere. Peter Bannon, who led the project, will exit the company as the remaining staff move over to other tech initiatives inside Tesla.
This comes amid a wave of recent departures, including about twenty employees who ditched Tesla to cofound a new AI enterprise called DensityAI. The startup, built by former Dojo head Ganesh Venkataramanan along with ex-Tesla engineers Bill Chang and Ben Floering, plans to make its debut soon. Their stated mission is to develop chips, software, and hardware for the data centers that power not just robotics, but also AI agents and automotive uses.
Tesla Shifts Course as Dojo Closes
These developments arrive at a critical moment for Tesla, with CEO Elon Musk urging investors to view the company not simply as an automaker, but a rising force in AI and robotics. Despite this pitch, Tesla’s limited robotaxi rollout in Austin was less than stellar, featuring cars piloted by humans in the passenger seat and generating troubling reports about inconsistent vehicle driving behavior.
Abandoning Dojo marks a clear pivot away from Musk’s previous promises. He often called Dojo the heart of Tesla’s quest for fully self-driving vehicles, touting its ability to handle mountains of video data and accelerate neural network training. Musk even made passing mention of Dojo on the firm’s latest earnings call, though detailed updates had slowed since last summer.
Analysts once thought Dojo could give Tesla a huge boost, with Morgan Stanley estimating it might someday add half a trillion dollars in market value by opening new revenue streams in robotaxis and software. Musk just last year told investors Tesla would “double down” on the project ahead of a major robotaxi reveal.
As fanfare around Dojo faded, Musk pivoted to a new vision: the Cortex supercluster, a huge AI system being constructed at Tesla’s Austin headquarters, now getting most of the executive’s attention.
Originally, the Dojo project united in-house chip development and custom computing. Tesla’s D1 chip, announced with some flair at the company’s 2021 AI Day, was meant to work alongside off-the-shelf Nvidia graphics chips. A more advanced D2 chip was also promised to fix data bottlenecks, but it never reached the same buzz.
With Dojo winding down, Tesla is leaning further into partnerships with established suppliers. This includes Nvidia and AMD for computing, and Samsung for manufacturing. Just a month ago, Tesla inked a major deal with Samsung to produce the AI6 inference chip, which could serve everything from the Optimus robot to powerful AI servers in data centers.
On a recent earnings call, Musk suggested that some activities may have overlapped. “Thinking about Dojo 3 and the AI6 inference chip, it seems like intuitively, we want to try to find convergence there, where it’s basically the same chip,” Musk said.
News of Dojo’s shutdown coincides with Tesla’s board offering Musk an enormous pay package worth $29 billion, a move designed to keep him focused on Tesla’s next AI chapter and not wandering toward outside ventures like xAI.
For a deeper dive into trends shaping AI data centers facing cost and power challenges, check out our latest analysis.
