Tool · verified 1 June 2026

GPU rental cost calculator

Drag the slider to model your monthly hours and see which cloud GPU provider gives you the cheapest bill. Real vendor-published rates, H100 through H200, RTX 4090 spot through Paperspace managed, plus a self-host line so you know when buying outright finally wins.

Run the numbers · interactive

Monthly cost across cloud GPU providers

RTX 4090

Consumer-tier · Stable Diffusion, hobby finetuning, sub-24GB workloads.

  1. Vast.ai RTX 4090 (marketplace best-bid) Marketplace spot. Variable, occasional preemption.
    $0.32/h — cheapest
  2. Runpod RTX 4090 (Community Cloud)
    $0.44/h — cheapest Try →
  3. Vultr RTX 4090
    $0.68/h — cheapest

L40S

Inference-tuned · 48 GB, throughput per dollar.

  1. Runpod L40S
    $1.09/h — cheapest Try →

A100 80GB

Workhorse · the default Llama 3 / Mistral fine-tune rig.

  1. Vast.ai A100 80GB (marketplace) Cheapest A100 hour but marketplace variability.
    $0.89/h — cheapest
  2. Lambda Labs A100 40GB (on-demand) 40 GB variant. Half the VRAM of the others.
    $1.1/h — cheapest
  3. Runpod A100 80GB (Community Cloud)
    $1.19/h — cheapest Try →
  4. Paperspace A100 (managed) Higher rate buys you a managed UX.
    $3.09/h — cheapest

H100 80GB

Production tier · most-rented enterprise card in 2026.

  1. Runpod H100 80GB (Secure Cloud) Production tier; usually available.
    $1.99/h — cheapest Try →
  2. Lambda H100 (reserved) Lower per-hour but requires committed term.
    $2.39/h — cheapest
  3. Vultr H100 80GB
    $2.79/h — cheapest

H200 141GB

Premium · 141 GB HBM3e, but listed ≠ rentable.

  1. Runpod H200 141GB (Secure, when listed) Frequently sold out; price listed not always rentable.
    $2.99/h — cheapest Try →
  2. Vultr H200 (announced)
    $3.3/h — cheapest
  3. Lambda H200 (on-demand)
    $3.5/h — cheapest

Self-host

Amortised over 3 years · the long-horizon crossover line.

  1. Buy H100 outright (amortised, 3yr) Assumes $30k H100 list, 36 months. Excludes power, cooling, downtime.
    $833/mo flat — cheapest

Vendor-published rates verified 2026-06-01. Excludes storage, egress, and region differences. See the methodology notes below.

How we picked these rates

Every per-hour number above is taken from the provider's published pricing page on 1 June 2026. We haven't run a workload on every instance; see our blog for those tests. These are listed rates, which matters because some of the H200 lines are listed but rarely actually rentable. Where that's true we've flagged it in the row note.

What we excluded

Storage, egress, snapshot retention, region multipliers, and any reserved-pricing commit. These move the bill more than people expect. A 20 TB month of egress on a "cheap" cloud can wipe out the per-hour saving entirely. If you want the full bill picture, start with our egress cost guide and the benchmark playbook.

How the self-host line works

We model an H100 purchase at $30,000 list, amortised over 36 months. That's the simplest possible breakdown. Power, cooling, downtime, and depreciation curves all move the real number. The line exists to show you the crossover band where on-demand renting is no longer the cheaper move (typically around 400 to 500 h/month at these prices).

When this gets stale

Cloud GPU pricing drifts. We review and bump the lastVerified date monthly; the date at the top of this page is the current pass. If you spot a stale rate, ping [email protected].

Want a deeper read?