In something like a mobile device, every watt counts. If it takes more energy to decode video on the tensor cores than it does to have a dedicated hardware block, you keep the hardware video decoder.