Tesla disbands Dojo team, shifts AI chip strategy
Tesla is reportedly disbanding the team responsible for its ambitious Dojo supercomputer, a move that signals a significant strategic shift away from in-house development for the core artificial intelligence powering its self-driving technology and Optimus humanoid robot. The dissolution of the Dojo unit is said to include the departure of its head, Peter Bannon.
Introduced by CEO Elon Musk in 2019, Dojo was envisioned as a “super powerful training computer” designed to process vast quantities of vehicle video data. Its primary purpose was to train the sophisticated AI models crucial for Tesla’s advanced driver-assistance systems. However, the company now appears to be reorienting its strategy, reportedly planning partnerships with industry giants Nvidia and AMD for computational resources, while entrusting chip manufacturing to Samsung. This aligns with Musk’s recent announcement last month of a substantial $16.5 billion deal with Samsung, aimed at developing Tesla’s next-generation AI6 chip. Responding to the reports, Musk affirmed on X that “Tesla AI5, AI6 and subsequent chips will be excellent for inference and at least pretty good for training. All effort is focused on that.” This statement suggests a renewed emphasis on chips optimized for both real-time decision-making (inference) and the iterative process of model refinement (training).
This strategic pivot comes at a critical juncture for Tesla, particularly as it rolls out its much-anticipated robotaxi service. The service recently debuted in Austin, Texas, albeit with a safety monitor present in the passenger seat. Its subsequent launch in San Francisco, however, notably places a human driver at the wheel, a decision that appears to contradict Musk’s earlier, emphatic promise of having “no one in the car” for the autonomous vehicles. The reliance on external partners for crucial AI infrastructure could be interpreted as a pragmatic response to the immense computational demands of fully autonomous driving and advanced robotics, potentially accelerating development or reducing the financial burden of maintaining a bespoke supercomputing initiative.
The reported disbandment of the Dojo team follows an earlier blow to the unit. Earlier this month, approximately 20 members of the Dojo team reportedly left Tesla to establish a new startup, DensityAI, which focuses on providing data center services across various industries, including automotive and robotics. Following this exodus, Tesla is said to have reallocated the remaining Dojo employees to other departments within the company.
The decision to transition from a custom-built supercomputer like Dojo to a more collaborative approach with established chipmakers marks a notable evolution in Tesla’s AI development philosophy. It suggests a pragmatic shift towards leveraging external expertise and resources, potentially streamlining its efforts in the intensely competitive fields of autonomous driving and advanced AI, even as it continues to develop its own custom silicon for specific applications.