Tesla shuts down Dojo AI supercomputer, shifts FSD strategy
Tesla is reportedly dismantling its ambitious Dojo supercomputer project, ending the automaker’s significant foray into developing proprietary chips for its driverless technology. The initiative’s lead, Peter Bannon, is departing the company, and the remaining team members will be reassigned to other data center and computing endeavors within Tesla, according to sources cited by Bloomberg.
This strategic pivot follows the departure of approximately 20 workers from Tesla, who subsequently founded their own artificial intelligence company, DensityAI. This new startup, reportedly emerging from stealth mode soon, is focused on building chips, hardware, and software designed to power AI data centers for applications spanning robotics, AI agents, and the automotive sector. DensityAI was co-founded by Ganesh Venkataramanan, formerly the head of Dojo, alongside ex-Tesla employees Bill Chang and Ben Floering.
The move comes at a crucial juncture for Tesla, particularly as CEO Elon Musk has actively sought to position the company as a leader in AI and robotics for shareholders. This re-branding push has occurred despite a limited robotaxi launch in Austin this past June, which utilized Model Y vehicles with a human safety driver in the passenger seat and reportedly experienced several instances of problematic driving behavior.
Tesla’s decision to discontinue Dojo marks a significant shift from Musk’s long-held vision. Since 2019, Musk has consistently touted Dojo as the cornerstone of Tesla’s AI ambitions and its pursuit of full self-driving capabilities, emphasizing its capacity to “process truly vast amounts of video data.” As recently as the company’s second-quarter earnings call, he mentioned Dojo, albeit briefly. In 2023, Morgan Stanley even projected that Dojo could add a staggering $500 billion to Tesla’s market value by unlocking new revenue streams through robotaxis and software services. Just last year, Musk indicated that Tesla’s AI team would “double down” on Dojo in anticipation of the company’s robotaxi reveal in October.
However, discussions surrounding Dojo began to wane around August 2024, when Musk started promoting “Cortex” instead—a “giant new AI training supercluster being built at Tesla HQ in Austin to solve real-world AI.” The Dojo project itself was multifaceted, encompassing both supercomputing infrastructure and in-house chip manufacturing. Tesla formally unveiled its D1 chip at its inaugural AI Day in 2021, with Venkataramanan presenting the chip, which was intended to work in conjunction with Nvidia’s GPUs to power the Dojo supercomputer. The automaker had also announced plans for a next-generation D2 chip, designed to mitigate potential information flow bottlenecks of its predecessor.
Now, sources indicate that Tesla intends to significantly increase its reliance on external technology partners. This includes a greater dependence on Nvidia for graphics processing units and AMD for other compute needs, alongside Samsung for chip manufacturing. Last month, Tesla reportedly finalized a substantial $16.5 billion deal with Samsung to produce its AI6 inference chips. This chip design is envisioned to be scalable, powering everything from Full Self-Driving and Tesla’s Optimus humanoid robots to high-performance AI training in data centers. Musk himself hinted at potential redundancies during the recent second-quarter earnings call, stating, “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.”
This news also surfaces as Tesla’s board is reportedly offering Musk a $29 billion pay package, aimed at retaining his focus on Tesla and advancing its AI efforts, rather than allowing him to be unduly sidetracked by his other ventures, such as the more pure-play AI startup xAI. Tesla has not yet publicly commented on the reports regarding Dojo’s shutdown.