Tesla Scraps Dojo Supercomputer, Focuses on AI5/AI6 Chips
Tesla CEO Elon Musk has confirmed the company’s decision to disband the team behind its ambitious Dojo AI training supercomputer project, just weeks after expressing expectations for a second Dojo cluster to operate at scale in 2026. Musk announced the pivot on X, the social media platform he owns, stating, “Once it became clear that all paths converged to AI6, I had to shut down Dojo and make some tough personnel choices, as Dojo 2 was now an evolutionary dead end.” He further suggested that a concept akin to “Dojo 3” might persist, redefined as a multitude of AI6 chips integrated onto a single board.
The original vision for Dojo was to create a powerful supercomputer to train the complex artificial intelligence models essential for Tesla’s self-driving technology and its humanoid robot initiatives. The initial iteration of Dojo utilized a combination of Nvidia GPUs and Tesla’s in-house designed D1 chips. Plans for a “Dojo 2” factory envisioned a second-generation D2 chip at its core, but it now appears the D2 chip, along with the broader Dojo project, has been shelved.
Tesla’s strategic shift now centers on its AI5 and AI6 chips, which are being manufactured by TSMC and Samsung respectively. The AI5 chip is primarily engineered to power FSD, Tesla’s driver assistance system. The more advanced AI6 chip is designed for a dual role: it will handle onboard inference, meaning it will process AI decisions directly within cars for self-driving and enable autonomous capabilities in humanoid robots, while also supporting large-scale AI training.
Musk elaborated on the rationale behind this consolidation, explaining that it no longer made sense for Tesla to allocate resources across two distinct AI chip designs. He asserted that the AI5, AI6, and subsequent chips will be “excellent for inference and at least pretty good for training,” emphasizing that all development efforts are now concentrated on this unified approach. He also noted that for a supercomputer cluster, integrating many AI5/AI6 chips onto a single board significantly reduces network cabling complexity and cost, potentially by orders of magnitude.
The concept of Dojo has been a fixture in Musk’s public statements since 2019, consistently presented as a foundational element for achieving full self-driving and commercializing humanoid robots. However, discussions around Dojo quieted around August 2024, when Musk began promoting “Cortex,” described as a “giant new AI training supercluster” under development at Tesla’s Austin headquarters. The current status of Cortex remains unclear. TechCrunch has sought clarification from Tesla regarding both Cortex and the fate of the Dojo facility in Buffalo, New York, where Tesla had reportedly invested $500 million.
This significant strategic realignment comes at a challenging period for Tesla, marked by declining electric vehicle sales and considerable brand reputation damage following Musk’s increasingly prominent forays into political discourse. Despite these headwinds, Musk has continued to assure investors of Tesla’s future in autonomy, even as the company’s limited robotaxi launch in Austin this past June faced reports of problematic driving behavior from the vehicles.