In the case of images I can just take an off-the-shelf pre-trained model like ResNet as a feature extractor, why should I use a cloud service for that? I don't

In the case of images I can just take an off-the-shelf pre-trained model like ResNet as a feature extractor, why should I use a cloud service for that? I don't quite get what the benefit of it is. Are your embeddings better? Well prove it then? In terms of transfer learning, fine-tuning convolutional layers would perform way better anyways?