Stop & Start
Stop a running GPU pod to stop paying for it without tearing it down, then start it later (on the same provider, or a different one) with your environment and data restored.
Provider coverage
Works the same way on every provider: stop and start need only a create call and a terminate call, which every provider we support implements. There's no storage precondition to check: Aquanode's own storage backs every pod's environment capture regardless of provider. Persistent volumes are the one thing that genuinely varies depending on the offer a pod is matched to.
What Stop actually does
Hitting Stop on a pod does three things in order: it captures the pod's Environment (the base image, packages, and startup script) and its Volume (/workspace: code, checkpoints, datasets) into their own storage, confirms both captures succeeded, then stops billing and closes the lease with the provider. The instance itself goes away. Stop is not a suspended, still-reserved VM sitting there waiting for you. What survives is the environment and the volume.
That ordering is deliberate: billing only stops after both captures are safely confirmed, so a failed capture keeps the pod Running with a loud error instead of silently losing your work along with the box.
What Start actually does
Start is not un-stopping the same machine. There is no machine left to un-stop. It rents a fresh instance, matched to the same GPU model and count as the one you stopped (or a specific offer you pick yourself, including one on a different provider), and restores your environment and volume onto it. If your environment has an app attached, that service is relaunched too, so you land back on a running ComfyUI or Jupyter instance rather than just a restored disk.
This is why a stopped pod is not billed while it sits stopped: there is no box behind it. And it's why starting onto a different provider is possible at all: the mechanism is the same one you'd use to move a workload by hand, just reached from the Start button instead of driven yourself.
The guarantee, and what it doesn't promise
The durable claim is this: once you stop a pod, your environment and volume are captured and can be restored onto a fresh box on any provider you choose. That's true regardless of which provider you stopped on: stop and start need only a create call and a terminate call, which every provider Aquanode supports implements, so there's no per-provider list to check before you stop something you're counting on getting back.
What it doesn't promise is that Start is instant or frictionless on the first try: it's renting a real box on a real provider, matched to a real offer, and that can take a moment or need you to point it at a different offer. There's no storage precondition to check first: Aquanode captures the environment regardless of provider. A pod with no Volume attached simply has nothing under /workspace to carry across the round trip.
Moving to another GPU without ever losing the box
Move to another GPU on a Running pod is a Stop followed by a Start on the new offer, run as one action. The machine is released only after both captures are confirmed, so a failed Start on the new GPU leaves the pod Stopped with its data intact rather than stuck between two machines.
When you'd take a workspace backup instead
A manual workspace backup is the guarantee that already runs continuously in the background while a pod is up, capturing your Volume on a tick between stops. Reach for a deliberate Stop instead of waiting on that tick when you want the clean checkpoint that only Stop gives you: it only happens once nothing is being written.
FAQ
Does a stopped pod keep billing?
No. Stop closes the provider lease along with capturing the environment and volume, so there is no running instance behind a stopped pod and nothing to bill for while it's stopped.
Can I start onto a different GPU or provider?
Yes. Start rents a fresh instance rather than restarting the stopped one, so you can point it at a different offer (including a different provider) and it restores your environment and volume onto that new box.
Which providers support Stop and Start?
All of them, in the sense that matters: your environment and volume are captured and can be restored onto a fresh box on any provider, and that doesn't depend on which provider you stopped on: stop needs only a create call and a terminate call, which every provider adapter implements. Persistent volumes are the one thing that genuinely varies depending on the offer a pod is matched to.
Is Start the same as un-stopping the same machine?
No. The original instance is gone once stopped. Start rents a new one and restores your environment and volume onto it, which is also what makes cross-provider Start possible in the first place.
Related reading: H100 pricing and specs, The best GPUs for AI, ranked, and Best GPU for LLM inference.
