In a surprising yet perhaps inevitable move, Elon Musk has officially pulled the plug on Tesla’s ambitious in-house supercomputing project known as Dojo. Once hailed as a revolutionary leap in AI computing, Dojo was envisioned to bring Tesla into a new era of self-reliant, high-performance AI infrastructure. But after years of setbacks, leadership shakeups, and growing dependence on external chipmakers, Tesla has now abandoned the Dojo effort, pivoting instead to strategic collaborations with Nvidia, Samsung, and AMD.
According to a Bloomberg report citing insiders, the Dojo project has been dissolved, its team disbanded, and its mission reassigned. Elon Musk himself confirmed the decision on his platform X, stating that “it didn’t make sense for Tesla to divide its resources and scale two different AI chips.” With this shift, Tesla is putting its future AI capabilities in the hands of proven industry players, marking a major course correction in its AI hardware ambitions.
 |
| Elon Musk Shuts Down Tesla’s Dojo Supercomputer Project |
What Was Dojo, and Why Did It Matter?
The Dojo supercomputer project was first announced in 2019 as part of Tesla’s broader vision to accelerate the development of autonomous driving and robotics. At its core, Dojo was designed to handle the massive quantities of data generated by Tesla vehicles—video feeds, sensor logs, driving patterns—and use it to train AI models at unprecedented speed and scale.
Tesla aimed to eliminate reliance on external AI hardware, particularly GPUs from Nvidia, by creating its own custom-built AI accelerator chips. These chips would then power the Dojo supercomputer—a massive cluster engineered to perform both AI training and inference, optimized specifically for Tesla’s use cases.
For Elon Musk, Dojo wasn’t just another infrastructure play; it was a cornerstone of Tesla’s AI-first strategy. It promised the kind of vertical integration he’s known for—building critical components in-house to gain tighter control over cost, performance, and innovation. However, the project never quite delivered on its bold promises.
A String of High-Profile Departures
Cracks in the Dojo dream began appearing early. Jim Keller, the legendary chip architect who originally spearheaded Tesla’s chip efforts, departed the company in 2018. He was succeeded by Ganesh Venkataramanan, another veteran engineer with stints at AMD and Apple. Venkataramanan carried the torch for several years but eventually left Tesla in 2023.
Peter Bannon, who had previously worked with Keller at Apple and took over leadership after Venkataramanan’s exit, is also leaving. These high-profile exits suggest mounting internal frustrations and possibly a lack of strategic clarity or technical progress. It's no coincidence that around 20 former Dojo team members are now reportedly working at DensityAI, a stealth-mode startup formed by ex-Tesla engineers. Early whispers suggest the new company may continue where Dojo left off—developing AI chips for robots and data centers.
Strategic Shift: Enter Nvidia, Samsung, and AMD
With the in-house effort now shelved, Tesla is turning to external tech partners to fuel its AI ambitions. Musk has already confirmed that Nvidia and AMD will supply Tesla’s next-generation computing infrastructure for AI training and inference tasks.
Nvidia, the current leader in AI computing, is expected to remain Tesla’s primary GPU provider for data centers and autonomous vehicle training. AMD will likely offer complementary solutions, particularly for edge inference chips or specific AI workflows.
However, the most significant pivot may be toward Samsung Electronics. In July, Tesla inked a massive $16.5 billion deal with Samsung to produce its next-generation AI chips. These include the upcoming AI5 and AI6 processors—designed by Tesla but manufactured by Samsung’s semiconductor division. According to Musk, Samsung’s cutting-edge chip fabrication plant in Taylor, Texas, will be the primary production site for these chips.
The AI5 chips are expected to go into production by the end of 2026, with AI6 to follow. These chips will be used not only in Tesla vehicles but also in its humanoid robots (Optimus) and high-performance data centers.
 |
| Elon Musk Shifts Focus to Nvidia, Samsung, and AMD for AI Hardware |
A New Chapter: “Dojo 3” in a Different Form?
Despite scrapping the original Dojo project, Musk hinted that its legacy might still live on—at least in spirit. On X, he suggested that combining thousands of Tesla-designed AI5 and AI6 chips in a large compute cluster could constitute “Dojo 3.” While it may not bear the original architecture or goals, the name lives on, symbolizing Tesla’s continued ambition to own key elements of the AI stack.
Still, this "Dojo 3" would fundamentally be different—it wouldn't be an in-house built supercomputer in the traditional sense. Instead, it would be a hybrid of Tesla-designed chips, manufactured externally, operating within infrastructure powered by other companies’ software and hardware platforms. It’s a practical compromise, one that prioritizes speed, scalability, and resource optimization over ideological purity.
The Real Reasons Behind Dojo’s Fall
While Tesla never publicly admitted that Dojo was failing, the writing was on the wall. In recent months, the company began buying significantly more computing capacity from Nvidia. Even as Musk publicly praised Dojo’s progress in June—claiming “Dojo 3 will be great”—Tesla was quietly ramping up purchases of Nvidia’s H100 GPUs, the very hardware Dojo was meant to replace.
Internal timelines were also slipping. Musk had previously claimed Dojo 2 would be operational at scale by 2025. With the project's cancellation, that milestone will now go unmet.
Moreover, Tesla’s broader internal restructuring likely played a role. Over the past year, the company has laid off thousands of employees and seen a wave of executive exits, many from its engineering and AI divisions. In this environment, continuing to sink resources into a slow-moving and high-risk project like Dojo became increasingly hard to justify.
What This Means for Tesla’s AI Future
While the end of Dojo may seem like a retreat, it’s more of a strategic refocusing. Tesla is doubling down on what it does best—vehicle design, robotics, and AI model development—while outsourcing the silicon-heavy lifting to those with better economies of scale.
Tesla’s integration with Musk’s other ventures also adds an interesting layer. The acquisition of social media platform X (formerly Twitter) for $33 billion, along with the growth of xAI and the launch of the Grok chatbot, hints at a broader vision: to bring advanced, conversational AI into Tesla’s vehicles and platforms.
Musk has already announced that Grok will be integrated into Tesla dashboards. With powerful chips like AI5 and AI6 on the way, such AI systems will be able to run locally in cars, reducing latency and increasing data privacy.
Final Thoughts
The end of the Dojo supercomputer marks the close of one chapter in Tesla’s AI journey, but it’s far from the end of the road. If anything, it’s a pivot toward a more agile, collaborative, and perhaps more realistic strategy.
Elon Musk is known for setting audacious goals, not all of which hit the mark. But when he shifts direction, it’s usually with a clear objective in mind. In this case, the goal is clear: to build world-class AI systems—faster, more reliably, and with the help of industry leaders who’ve already solved the hardest hardware problems.
Dojo may be gone, but Tesla’s AI ambitions are far from over. They’re just getting a new kind of fuel.
0 Comments