Elon Musk, CEO of Tesla, is paying over $1 billion to develop an A.I. supercomputer called Dojo to teach his “Full Self-Driving” (FSD) software and Optimus humanoid robots. To teach both FSD software and Optimus robots, the business need Nvidia’s powerful A100 tensor core GPU clusters. Because Tesla has a huge need for A.I. CPUs, Musk believes Dojo will be more efficient than purchasing Nvidia chips off the market. Tesla intends to attain 100 exaFLOPS in-house computing capabilities by the end of next year, placing it among the top five suppliers globally.
Musk is ready to forego profit in exchange for volume because he believes his cars will drive themselves without human intervention. Instead of depending on several redundant sensor systems, Musk wants his cars to be sophisticated enough to drive themselves anywhere using only camera data. Musk is utilizing 300 million miles of real-world driving data taken everyday from customers’ automobiles to train his autos. This information will be input into Dojo, which is meant to drastically lower the cost of neural network training by utilizing unique silicon created internally at Tesla. Musk believes that in the long run, autonomy will boost car sales volume through the roof.
