Haven’t done this in earnest yet, so just a few links.
Data science VMs
ML workbench quickstart, which i think is now the same thing as ML Studio. (?)
They offer FPGAs! And FPGAs are hip!
GPU VMs. If you use VS Code, they offer integration with that.
For intermittent users, you would want ideally to use a serverless style, which I think they call Functions, although it sounds like they are not architected for e.g. giant GPU processes.