Does autoscaling to zero prompt rebuild every time it receives a new request?

For HF Spaces, if a space is configured to sleep after any amount of time, it will rebuild the image when it is accessed after sleeping. Do Inference Endpoints have a similar behavior after autoscaling to zero? I’m concerned about this because docker build is extremely slow if the image contains multiple large files. I want to save costs when no one is using my tool. However, my tool uses several very large Lightning checkpoint files, so running it on a HF space with sleep seems impractical, as every time it wakes from sleep it will build for an hour or so.