Modal

Modal runs Python functions in the cloud with minimal setup, and the GPU support is legitimately good. You decorate a function, specify what hardware you want (T4 through H100), and it handles the containers and scaling. The sweet spot is ML inference and batch jobs where you need GPUs but don't want to manage infrastructure. Setup is clean: install the package, authenticate once, and you're deploying. The autoscaling actually works, you only pay for what you use, and new accounts get $30 monthly credits. If you're running transformers or doing parallel data processing and tired of configuring cloud instances, this removes most of the friction.