Hello everyone, I am Xiao Luo. Today we will talk about Tesla.
Tesla recently announced the disbandment of its self-developed training chip project, the Dojo team, with relevant leaders also resigning one after another. This news has attracted widespread attention in the industry.The Dojo project was once highly anticipated, aiming to achieve training performance that surpasses existing GPU platforms through self-developed chips and systems.
So, what is the Dojo project?
The Dojo project was first proposed in 2019, going through a phase of high-profile launch and technological breakthroughs. Its core technologies include the self-developed D1 chip and Training Tile module, aiming to build an ExaPOD-level training system. However, over time, the project gradually exposed issues such as insufficient ecosystem development and commercialization difficulties.
Meanwhile, Tesla has launched a training cluster based on NVIDIA’s H100 in Austin, continuously expanding to tens of thousands of GPUs,ultimately replacing Dojo as the main platform.
The core failure of the Dojo project lies in the lack of an ecosystem. In contrast, NVIDIA not only leads in hardware performance but also builds a complete ecosystem through software systems like CUDA and cuDNN, as well as networking technologies like NVLink and NVSwitch.This ability for software-hardware synergy makes NVIDIA’s products not only powerful in performance but also quickly accepted and applied by the industry.
Looking back at Dojo, despite not falling behind in technical indicators, the lack of industry-recognized ecosystem supportled to its inability to form scalable commercial applications.
As we all know, the time and financial costs of developing self-developed training chips are extremely high. Tesla has made significant investments in chip development and system integration but ultimately failed to keep pace with market rhythms. In contrast, NVIDIA occupies the main chip market by integrating GPUs, networks, and software into a directly deployable AI factory.
The end of the Dojo project reminds us once again thatthe competition in AI chips is also a competition in technology, not just a technical contest, but also requires a higher strategic positioning in business.It is extremely challenging for companies to complete the entire chain of development from chip research and development to ecosystem building on their own; collaboration is key.
Tesla’s choice is pragmatic, while NVIDIA’s victory is a systemic victory.In the era of AI infrastructure, speed and ecosystem are everything.
If this article has inspired you, remember to give it a thumbs up and share it with more friends. Your support is my greatest motivation to continue sharing valuable content! Thank you for reading, and see you next time!