Maybe the DO or Paperspace team can build something like RunPod. They support containers and per-minute billing. Or like replicate.com but find a way to have a

Maybe the DO or Paperspace team can build something like RunPod. They support containers and per-minute billing. Or like replicate.com but find a way to have a model sleep without being completely unloaded. Maybe a good business model could be a system built around LoRAs or the equivalent for LLMs. That way you could share a lot of instances running the same containers but just swap out the adaptation data quickly for different customers.