Why did Tesla give up its independent chips and supercomputers?
The AI supercomputer "Dojo" project that Elon Musk has emphasized over the past few years finally came to an end. In August 2025, Tesla officially announced the disbandment of the Dojo team and completely terminated the project. Just a year ago, Tesla had expressed confidence that it would commercialize in 2026 through the second-generation Dojo, but just a few weeks later, it defined it as an "evolutionary dead end." Dojo was once a landmark project that symbolized Tesla's transformation into an AI company through surpassing a pure electric car manufacturer. Now, how should we interpret this "grave of AI"?
Dojo is a supercomputer designed by Tesla to train its own autonomous driving neural network. Its core lies in Tesla's self-developed chip "D1/D2". The original intention of this concept is to create a system that does not rely on NVIDIA GPUs and has faster computing speed and low latency architecture.
Musk once painted a grand blueprint through Dojo: Tesla will use Dojo to complete "vision-based autonomous driving" and apply it to autonomous taxis (Robotaxi) and humanoid robots (Optimus). By storing and processing massive driving image data, artificial intelligence that simulates human vision is trained - this technological path itself is extremely attractive.
But the reality is contrary to its expectations. Tesla failed to directly associate the achievements of autonomous driving with Dojo, and its chip performance has never been able to catch up with the evolutionary speed of general-purpose GPUs such as Nvidia H100 and H200. More importantly, the mainstream software in the AI ecosystem is optimized for GPU, which has also become a major obstacle to Dojo's development. In the end, Dojo's persuasiveness at the cost, manpower and technical levels gradually disappeared.
Dojo's death cannot be regarded as a failure of a project only, it also reveals multiple key issues:
First, it proves that "realizing complete technological autonomy through self-developed chips and self-developed supercomputers" is a very risky strategy. Although Tesla dares to try to independently develop chips and infrastructure, its resources and ecosystem limitations are fully exposed in the process of catching up with the amazing development speed of the AI semiconductor industry.
Secondly, it highlights the chain reaction of talent loss. After Dojo was disbanded, some core personnel formed a startup called "DensityAI", causing Tesla's internal technical assets and experience to flow outside. This phenomenon fully demonstrates that special technical projects are extremely dependent on talents, and the loss of talents may even directly lead to the collapse of the project.
Third, it reflects Musk's decision-making style characteristics. Until 2024, he was still emphasizing the importance of Dojo, but he changed direction completely after just one year, signing a development contract for the next generation of AI6 chips with Samsung. The end of Dojo is essentially a signal of Tesla's strategic transformation from "independent research and development" to "utilizing partner resources".
Tesla: Is it an AI company or an automotive company?
Although Dojo has terminated, Tesla's AI ambitions have not been extinguished. At present, the company is still cooperating with companies such as Nvidia, AMD, and Samsung to promote the expansion of the supercomputer "Cortex" based on AI6 chips. In fact, the training work for the latest version of fully autonomous driving (FSD) is already under the responsibility of Cortex.
Even so, the dissolution of Dojo still leaves a key problem: Although Tesla calls itself an "AI enterprise", its positioning has fallen into turmoil again after its most iconic AI project died in the middle of the journey. The slowdown in electric vehicle sales growth, the limited launch of autonomous taxi services, and the retreat of AI chip strategy are all sounding alarms for investors and the entire industry, allowing people to face the cold reality.
While Dojo ended in failure, it will be a classic case of how bold Tesla has tried in the high-risk areas of autonomous driving and AI. The key now is that this failure is just an end point or a sacrifice for a larger transformation.
