The cheapest GPU cloud, measured
GPU cloud prices for the same H100 silicon ranged from $2.07/hr (Vast.ai marketplace) to $4.59/hr (Lambda) and $6.16/hr (CoreWeave) on 2026-11-11, a 3x spread on one chip. The cheapest tier is usually preemptible or marketplace capacity, and the cheapest list price is not always buyable. rent.dev routes a job to the cheapest provider that actually has stock — one endpoint, one bill.
rent.dev — one endpoint for any compute an agent needs.
Questions
- Which GPU cloud is cheapest?
- On 2026-11-11 the lowest live H100 SXM offer was $2.07/hr on Vast.ai's marketplace, with RunPod Community at $2.69/hr and Daytona preemptible at $2.27/hr, against $3.95/hr on Modal and $4.59/hr on Lambda. The cheapest tier is interruptible capacity, so it suits jobs that checkpoint.
- Why do GPU cloud prices vary so much?
- The same silicon is sold at different tiers: marketplace hosts, preemptible pools, on-demand and committed contracts are priced independently, so a single H100-hour varies by 3x or more across providers on the same day.
- How do I rent a GPU for an hour?
- rent.dev is one endpoint for GPU and compute rental: an agent posts a job spec and it is placed on the cheapest provider with stock, billed once.