Tesla has made a bold and unexpected step. The business has officially disbanded its Dojo supercomputer team, signaling the end of a years-long endeavor to develop a bespoke AI training system. Dojo was supposed to be Tesla’s strong internal platform for training self-driving and robotics AI models with video footage from millions of Tesla vehicles.
The announcement comes after approximately 20 Dojo developers left to form DensityAI, a new firm focused on building AI chips and technology for robotics and data centers. While some fans are surprised, Tesla claims it is just spending resources on projects that correspond with its larger mission.
What Was Project Dojo?
Project Dojo was first announced during Tesla’s AI Day in 2021. Its goal was straightforward but ambitious: handle enormous volumes of real-world video data quickly so Tesla’s AI could learn faster.
The system was powered by Tesla’s D1 processor, with an improved D2 version planned to increase speed and address performance concerns. Dojo was hailed as a game-changer that might impact not only Tesla’s automobiles but also the AI industry.
The technology was designed to work at unprecedented scale, processing video data directly from Tesla’s global fleet rather than depending on standard cloud providers. One of its distinguishing features was the use of wafer-scale integration, which enabled Tesla to fit significantly more processing units onto a single piece of silicon than conventional chips. This strategy offered lower latency, more bandwidth, and lower energy consumption for AI training workloads.
Dojo’s architecture was also modular, allowing Tesla to enhance its capabilities simply by adding additional interconnected tiles, a design choice that piqued the eye of the larger AI research community.
Why Tesla Pulled the Plug
Enthusiasm began fading in mid-2024. That’s when CEO Elon Musk started promoting Cortex, a new AI supercluster being built at Tesla’s Austin HQ. Musk wanted a unified chip strategy rather than splitting resources between two different designs.
On X, Musk explained:
“It doesn’t make sense for Tesla to divide its resources and scale two quite different AI chip designs. The Tesla AI5, AI6, and subsequent chips will be excellent for inference and at least pretty good for training. All effort is focused on that.”
In short, Tesla wants one architecture that works for both training and running AI models, instead of juggling multiple systems.
The Shift to AI5 and AI6 Chips
Tesla’s decision to close the Dojo chapter is not the end of its AI hardware ambitions; rather, it is a strategic move. The emphasis is now on AI5 and AI6 chips, which are intended to perform both inference (running AI models in real time) and some level of training. Inference is crucial for products that require speedy decision-making, such as self-driving cars and robotics.
By focusing on these chips, Tesla hopes to establish a unified architecture that can be used for numerous purposes rather than separating resources across completely different systems. This strategy eliminates complexity, accelerates production, and decreases costs.
AI5 chips are expected to be available by 2026, followed by AI6 processors under a $16.5 billion production agreement with Samsung that extends until 2033. These processors will not only power Tesla’s Full Self-Driving platform but also play an important role in the company’s future robotics initiatives, including the Optimus humanoid robot and other AI-driven projects.
What This Means for Tesla’s Future
The end of Dojo does not mean that Tesla is abandoning AI; rather, it signifies that Tesla is changing its approach to it. Instead of pursuing a specialized training chip that requires a significant expenditure and a distinct architecture, Tesla is simplifying its plan. By focusing on AI5 and AI6 chips, the company can concentrate on fewer designs that nevertheless provide great computational power while being easy to expand.
The inclination offers practical benefits. First, it minimizes production complexity. With a single, unified chip architecture, Tesla can create both training and inference systems using the same technology, saving manufacturing costs and simplifying maintenance. Second, it speeds up product development. Tesla’s technical teams can work more quickly on incorporating AI into products like the Optimus humanoid robot, the Cybercab, and future Full Self-Driving updates since they are not distracted by supporting two completely separate chip platforms.
Partnering for Strength
Partnering with giants like Samsung, Nvidia, and AMD means Tesla can tap into their deep manufacturing expertise and proven hardware performance. This move assures that Tesla will not bear the full risk of building new processors from scratch—a lesson learnt after years of investment in Dojo.
While some analysts assessed Dojo’s potential at $500 billion, Musk’s choice shows that Tesla prioritizes execution over status. It’s less about having the most distinctive processor on the market and more about integrating dependable, scalable AI into Tesla’s products as soon as possible.
This move is also consistent with a broader trend in the technology industry: consolidation. Companies are gradually building fewer but more adaptable processors capable of handling diverse workloads. For Tesla, this may mean shorter time-to-market for AI-powered breakthroughs, more competitive pricing for its technologies, and a stronger position in the fight for autonomous mobility and robotics supremacy.
Musk has promised that the next-gen AI5 chips will arrive by the end of 2026. AI6 chips, made by Samsung, will follow. These will be central to Tesla’s push into autonomous driving and humanoid robotics.
While the Dojo name might fade, the vision behind it—making Tesla’s AI smarter and faster—isn’t going anywhere. As Musk put it:
“One could call that Dojo 3, I suppose… It’s real good.”
Tesla’s AI future is still alive—just on a different path.
