When does serverless get too pricey or complicated for stuff that runs a long time?

Okay, so I'm building a project and it involves some pretty long, CPU-heavy stuff running in the background, like encoding videos or doing complicated data analysis. We've been checking out serverless options, things like AWS Lambda or Azure Functions, because they're great for scaling up and you don't have to worry about managing servers. And yeah, for quick, event-triggered tasks, they're awesome. But I'm starting to think about the costs and how complicated the architecture might get for processes that could run for several minutes, maybe even an hour. I mean, these functions could easily hit memory limits or their maximum execution times. So, at what point does it just make more sense, cost-wise or just simpler, to use good old containerized services, like ECS or Kubernetes on EC2 or similar? Is it just about figuring out the cost per use versus having resources always running? Or are there other things, like easier state management or debugging, that would make you lean towards VMs or containers for these kinds of jobs?