For your local development environment, I recommend using Windows. You can't effectively use Mac because, due to their use of AMD rather than Nvidia, GPU-based ML frameworks can't optimize to Mac. You could use Ubuntu Desktop, but it's not for the weak-of-heart due to various software compatibilities; and when it comes to laptops, drivers aren't as up-to-snuff as Windows drivers (especially with battery lifetime & wifi). So use Windows.

Use Docker. You'll need it eventually for cloud-deploy anyway, so might as well start now and gain the dev environment benefits. To use Docker with GPU on Windows, follow these instructions.

• Setup Windows Dev Channel and WSL2 with nvidia-docker support. Instructions here
  • Alternatively: Nvidia's guide for CUDA on WSL2
• Odds-and-ends with WSL2 & Docker here
• Create a Dockerfile inheriting from a popular pre-vetted ML setups like nvidia/cuda:10.1-cudnn7-runtime-ubuntu18.04, huggingface/transformers-gpu, or Gnothi's ml-tools kitchen-sink.

Details on buying / building a good PC desktop or laptop for machine learning. Details on front-end stack + cloud-hosting and ML cloud-hosting.