Because maybe there's some advantage to be had by reducing dependencies when you can afford to? And maybe they value SOC expertise, since they do other tech besides cars?
And when they outsource it people will say they're just focusing on core competencies. The great thing is businesses can use a magic 8-ball to make business decisions and someone will laud the results as a good move.
Vertical integration has many benefits. It’s the same reason Apple started making their own chips and ditched intel. The SOCs will be cheaper and better optimized for Tesla’s use case.
Every Nvidia GPU I’m aware of ships with both CUDA cores and Tensor cores. For a pure ML application those CUDA cores are almost useless. Tesla would be paying Nvidia’s premium for transistors that aren’t optimized for ML inference.
A Tesla designed chip can be 100% dedicated to tensor multiplication. You’re paying less per transistor and every transistor is utilized to the fullest extent.
And why would anyone bother with custom ML SOC when you can just partner with Nvidia like Rivian, Mercedes etc.