Workers at those providers do not run _at the edge_ for each request. They lean on AWS Lambda to pre-generate or render assets, but you can't reliably run a function there on every user-facing request without a noticeable latency penalty.