GPU Rental Prices
A daily ledger of cloud GPU rental prices: what it actually costs, per GPU per hour, to rent an H100, H200, B200, A100, MI300X, RTX 4090/5090, L40S, GH200 and other accelerators from 22 cloud providers. Every price row carries its own provenance: the provider page or API it was fetched from and the exact fetch timestamp.
The dataset is append-only. A new snapshot is recorded every day; history starts 2026-07-05 and cannot be backfilled, which is the reason it is recorded at all.
Providers currently tracked: AWS, Azure, CoreWeave, Crusoe, DataCrunch, fal, Fireworks, Hyperstack, Jarvis Labs, Lambda, Massed Compute, Modal, Nebius, OVHcloud, Replicate, RunPod, Salad, Spheron, TensorDock, Together, Voltage Park, Vultr.
Files
| File | Contents |
|---|---|
data/latest.json |
The most recent verified snapshot (all offers plus per-provider fetch status). |
data/snapshots/YYYY-MM-DD.json |
One snapshot per day, append-only. This repo keeps a rolling window of recent daily snapshots; the full history lives on gpurentalprices.com. |
Schema
Each snapshot file is a JSON object:
| Field | Type | Description |
|---|---|---|
date |
string | Snapshot date, YYYY-MM-DD. |
generated_at |
string | ISO 8601 timestamp when the snapshot was generated. |
offers |
array | One record per provider x GPU x pricing tier, see below. |
providers |
object | Per-provider fetch status: ok, last_verified, stale, and error when a fetch failed. |
meta |
object | Snapshot totals: provider count, ok/stale counts, unrecognized SKUs (present in latest.json). |
Each record in offers[]:
| Field | Type | Description |
|---|---|---|
provider |
string | Provider slug (e.g. runpod, lambda, aws). |
gpu |
string | Normalized GPU id (e.g. h100-sxm, a100-pcie-80, mi300x, rtx-4090). |
vram_gb |
number | GPU memory in GB. |
usd_hr |
number | Price in USD per GPU per hour. Excludes storage, egress and networking. |
kind |
string | Pricing tier: on-demand, secure, community, spot, or serverless. |
source_url |
string | The provider pricing page or API the price was fetched from. |
fetched_at |
string | ISO 8601 timestamp of the fetch. |
min_gpus |
number | Optional. Minimum GPU count for the listed price, when the provider only sells in multiples. |
Collection and update cadence
Once per day, each provider's own pricing source is fetched: a public API where one exists, the official pricing page otherwise. Prices are normalized to USD per GPU per hour. When a fetch fails, that provider's rows are carried forward and flagged with their last-verified date rather than silently presented as current. Full method: gpurentalprices.com/methodology.
This Hugging Face dataset is synced automatically after each daily update.
Links
- Canonical site (sortable table, history charts, per-cell source and timestamp): gpurentalprices.com
- Live JSON API: gpurentalprices.com/api/latest.json
- GitHub mirror: github.com/adriannutiu/gpu-rental-prices
- Methodology: gpurentalprices.com/methodology
License
CC BY 4.0. Free to use for dashboards, notebooks, research, price alerts, anything. Attribution: "GPU rental price data by gpurentalprices.com, CC BY 4.0."
Citation
DOI: pending
APA:
GPU Rental Prices. (2026). GPU rental price index [Data set]. gpurentalprices.com. https://gpurentalprices.com/data
BibTeX:
@misc{gpurentalprices_2026,
title = {GPU Rental Price Index},
author = {{GPU Rental Prices}},
year = {2026},
howpublished = {\url{https://gpurentalprices.com/data}}
}
- Downloads last month
- 493