GPU EconomyCloud GPU prices, read every hour
updated
H100 index$3.5426 providersWidestV10015.9×Mover, 7dB300−1.4%Coverage709 quotes63 providersReads, 24 h17,34826 runs

GPU cloud · 4 GPU models tracked

Replicate

$0.81–$5.49per GPU-hour, the cheapest row to the dearest

Models run on hardware billed by the second - public models for the seconds a request runs, private ones for all the time their instances are up - shown here as an hour; each price is an instance with its CPU and memory. H200 and 4- or 8-GPU machines are sold by contract and are not listed.

As of , Replicate lists 4 GPU models we track, from $0.81 per GPU-hour (NVIDIA T4, on-demand) to $5.49 (NVIDIA H100 80GB (form factor not stated), on-demand).

Every rate ↓ · Compare with 32 providers ↓

  • Cheapest row$0.81NVIDIA T4 · on-demand
  • A100 80GB$5.04cheapest tier listed
  • L40S$3.51cheapest tier listed
Websitereplicate.com/
Pricing pagereplicate.com/pricing
Regions and tiersNo region or tier named in the rates below: one global rate card
Models tracked4

Price by GPU

As of , per GPU-hour, cheapest tier first named:

All listed prices

One table per price type, never mixed. Every rate normalized to one GPU for one hour.

On-demand

Listed pay-as-you-go rates at Replicate.
GPUUSD / GPU-hour7-day changeTypeRegionMin. commitmentSource, capturedDeploy
NVIDIA T4$0.81—on-demand——Price page

unchanged since first read, Sep 23, 2026
Deploy ↗
NVIDIA L40S$3.51—on-demand——Price page

unchanged since first read, Sep 23, 2026
Deploy ↗
NVIDIA A100 80GB$5.04—on-demand——Price page

unchanged since first read, Sep 23, 2026
Deploy ↗
NVIDIA H100 80GB (form factor not stated)$5.49—on-demand——Price page

unchanged since first read, Sep 23, 2026
Deploy ↗

Every rate on this page as data: /providers/replicate/prices.json — the same fields as prices.json, narrowed to this provider. CC BY 4.0, no key, no rate limit. On your own page: embed this table (/embed/providers/replicate, an iframe, no script).

Compare Replicate with

Every provider that lists at least two of the same GPUs on-demand, most shared GPUs first; each opens the side-by-side page.

Data updated: