RTX 4090s from $0.30/hr, billed by the second, drivable by AI agents
I'm Aaryn, founder and principal engineer at Xcelsior Compute in London, Ontario. We run a distributed GPU marketplace where people rent verified hardware, like RTX 4090s, billed by the second from $0.30/hr. You can also deploy open models as token-billed serverless endpoints that scale to zero. Hosts pass seven verification checks before they take jobs.
The part I care most about is making the platform usable by AI agents. We ship an MCP server, and the in-dashboard assistant asks before any action that spends money. Spot jobs lock at the posted rate when they're assigned, so there's no auction. I don't have usage numbers I'd put in a post yet. What I'm still working out is how to explain that pricing model to people who have only used fixed-price clouds.
The most useful thing you could offer is feedback from anyone who has rented GPUs on Vast.ai or RunPod and hit a billing surprise or an underperforming host. Tell me what you would need to see verified before trusting a host with a real job.






Comments1
Welcome Aaryn! Per-second billing on 4090s makes a lot of sense for small teams running short jobs. Curious how you verify the hardware people list on the marketplace.
1 comment
Welcome Aaryn! Per-second billing on 4090s makes a lot of sense for small teams running short jobs. Curious how you verify the hardware people list on the marketplace.