Tesla’s Dojo supercomputer — once presented as a cornerstone of the company’s AI ambitions — has been abruptly scrapped. CEO Elon Musk confirmed the closure, but the explanation he offered has left industry watchers unconvinced.
The move is a sharp departure from Musk’s recent public enthusiasm for the project. Just last month, he described Dojo’s second iteration as “good” and suggested that the third generation would be “great.” The system was promoted as Tesla’s custom-built platform for training AI models to power its self-driving software, potentially reducing reliance on Nvidia’s dominant hardware.
Bloomberg’s Account: A Mass Departure
The first signs of trouble surfaced in a Bloomberg report, which cited unnamed sources claiming that around 20 key engineers left Tesla’s Dojo team to join a new startup, DensityAI. Among them was Peter Bannon, a senior chip architect responsible for both Tesla’s in-vehicle AI6 inference chips and Dojo’s training chips.
Bloomberg’s sources suggested the departures critically weakened Tesla’s ability to advance Dojo and compete with entrenched players in the AI hardware market, making the project’s continuation difficult to justify.
Musk’s Reasoning: The AI6 Factor
Musk did not dismiss Bloomberg’s reporting but instead pointed to a different cause. He claimed that Tesla’s AI6 chip — developed for inference computing inside vehicles — was so capable that it rendered Dojo’s specialized training chips unnecessary. He described Dojo’s second version as an “evolutionary dead end,” framing the AI6 as a unified solution for future computing needs.
In Musk’s view, this made it logical to wind down the Dojo effort and reallocate resources. He implied that what he called “Dojo 3” would effectively live on as a cluster of AI6 chips.
Why Experts Are Skeptical
Industry analysts have questioned whether a single chip design can truly serve both training and inference purposes at a world-class level. The two workloads have different demands: training favors high numerical precision, while inference prioritizes low latency and energy efficiency.
It’s not impossible to use the same hardware for both, but doing so often involves compromises. If Tesla’s AI6 could outperform specialized training chips, it would be a rare feat — and one not yet demonstrated publicly.
The exit of Bannon adds another layer of doubt. If AI6 was as groundbreaking as Musk claims, retaining the architect behind it would seem an obvious priority. His departure lends weight to Bloomberg’s account of an internal talent drain influencing the decision.
An Abrupt Turnaround
The contrast between Musk’s upbeat July comments and the sudden dismantling of Dojo is stark. Just weeks earlier, he had spoken of scaling Dojo 2 within a year and envisioned an “AI factory” in production by late 2025. He also discussed the possibility of convergence between Dojo 3 and Tesla’s vehicle and robotics chips.
The quick reversal raises the possibility that the shutdown was less a planned shift in strategy and more a reaction to unforeseen challenges — whether technical, competitive, or personnel-related.
A Risky Bet That Didn’t Pay Off
Dojo was an ambitious attempt to reduce Tesla’s dependence on Nvidia’s GPUs, a move that could have lowered costs and improved control over AI development. But building high-performance training hardware is a technically demanding and capital-intensive endeavor, especially against industry leaders with decades of experience.
The loss of seasoned engineers to DensityAI, coupled with the inherent difficulty of the task, may have pushed Tesla to conclude that sticking with proven external suppliers was the more practical route.
Familiar Patterns in Musk’s Projects
Musk has a track record of making bold promises about transformative technology, only to alter course or delay delivery. Predictions about full self-driving cars, robotaxis, and humanoid robots have often slipped years past their initial timelines.
This pattern — ambitious announcements followed by abrupt changes — has inspired both loyal supporters and frustrated critics. While some see it as necessary risk-taking in pursuit of breakthroughs, others view it as overpromising that risks damaging credibility.
Investor and Community Reaction
The shutdown has divided Tesla’s online community. Some long-term investors expressed disappointment, arguing that Musk’s recent public statements about Dojo painted an overly optimistic picture given how close the project was to cancellation.
Others suggested that focusing on the AI6 could be a smart move if it truly enables Tesla to streamline its hardware approach across vehicles and robotics. However, without concrete performance data, that remains speculative.
Without Dojo, Tesla will likely continue to rely heavily on Nvidia’s hardware for AI training, competing with other sectors also hungry for advanced chips. While this approach offers stability, it may slow Tesla’s ability to scale its self-driving technology development as quickly as it had hoped.
If AI6 proves capable of handling both training and inference effectively, it could mark a unique step forward for Tesla’s AI infrastructure. But if the chip’s capabilities have been overstated, the decision to end Dojo could be seen as a significant strategic setback.
