Google Tensor has broke through Nvidia's almost monopolistic position on AI story and chips.
But I am wondering and asking if someone with more knowledge can explain to me if Google Tensor can supplement or replace nvidia chips.
The major reason why AMD could not directly compete with Nvidia was that CUDA was created by Nvidia decades ago and all AI developers use CUDA and its library. Even though AMD hardware can compete with Nvidia hardware, CUDA not officially supporting AMD chips made it undesirable for the industry. AMD tried to help it with unofficial support of CUDA as their own solution is not widely adopted.
Afaik Google Tensor cannot use CUDA officially either. Is this a problem for Tensor and for AI devs?