GPU Rental Price History

Weekly price history for 103 GPU models across 82 cloud providers.

Free with attribution · CC BY 4.0 · updated daily

GPU models
103
with price history
Providers
82
in the price history
Weeks
53
trailing, per model
Updated
21 Sep 2026
once a day

Download

One row per GPU model, billing type and week. Prices are per GPU per hour. The latest week may be partial and can change.

# Every GPU model: the latest week's on-demand medians, cheapest first
import pandas as pd

prices = pd.read_csv("https://getdeploying.com/dataset/gpu-prices/weekly.csv")
on_demand = prices[prices.billing_type == "ON_DEMAND"]
latest = on_demand[on_demand.date == on_demand.date.max()]
latest.sort_values("median_price")[["gpu_slug", "median_price", "provider_count"]]
# One GPU model per file: chart two against each other
import pandas as pd

h100 = pd.read_csv("https://getdeploying.com/dataset/gpu-prices/nvidia-h100.csv")
rtx = pd.read_csv("https://getdeploying.com/dataset/gpu-prices/nvidia-rtx-pro-6000.csv")
both = pd.concat([h100, rtx])
both = both[both.billing_type == "ON_DEMAND"]
both.pivot(index="date", columns="gpu_slug", values="median_price").plot()

Per-provider rows, daily data and longer history are in the API.

Every GPU model

One file, keyed by gpu_slug. 15,078 rows.

Download CSV · 981 KB JSON

One GPU model

One file per GPU model. Swap in the slug.

/dataset/gpu-prices/<slug>.csv
/dataset/gpu-prices/<slug>.json

Preview

weekly.csv · 6 of 15,078 rows

gpu_slug date billing_type reservation_months min_price max_price median_price provider_median_price provider_count offering_count
nvidia-h100 2026-09-21 ON_DEMAND 1.3 12.29 3.19 3.3475 39 139
nvidia-b300 2026-09-21 ON_DEMAND 6.5 17.802 7.945 7.89 14 16
nvidia-b200 2026-09-21 ON_DEMAND 3.75 16.2744 6.2521 6.2511 18 31
nvidia-rtx-pro-6000 2026-09-21 ON_DEMAND 0.66 8.294 2.1233 2.195 40 155
nvidia-h200 2026-09-21 ON_DEMAND 2.09 13.78 4.3435 4.37 35 91
nvidia-rtx-5090 2026-09-21 ON_DEMAND 0.35 2.52 0.69 0.69 15 119

Fields

10 columns, the same names in the CSV and the JSON.

Field Type Definition
gpu_slug text The GPU model's slug, as in its page URL (supported slugs)
date date, ISO 8601 Monday of the week the row describes
billing_type text How the listing is billed
reservation_months integer, or empty Term of a reservation, in months
min_price USD / GPU / hr Cheapest listing
max_price USD / GPU / hr Most expensive listing
median_price USD / GPU / hr Median over all listings
provider_median_price USD / GPU / hr Median of each provider's own median, so every provider counts once
provider_count integer Providers with a priced listing not marked sold out
offering_count integer Priced listings not marked sold out

License and attribution

CC BY 4.0: Credit GetDeploying and link to the license. 

GetDeploying, GPU rental price history, https://getdeploying.com/gpus, CC BY 4.0
GetDeploying (2026). GPU rental price history [dataset]. https://getdeploying.com/dataset/gpu-prices. Accessed September 21, 2026.

The license covers the published dataset files only and implies no endorsement. The site and the API stay under the terms.

Common questions

How is a weekly row calculated?

Prices are captured once a day; a weekly row is the median of that week's daily values, column by column, so one bad capture is outvoted. No smoothing, no interpolation: a week with no capture is absent, and the latest week can change until it ends.

Which listings are counted?

Every listing a provider prices and does not mark sold out, including unknown and waitlisted availability, which is most of them. These are advertised prices, not checked as rentable. A median moves when providers join or leave as well as when anyone reprices; provider_count is how to tell.

How often does it update?

Once a day, within about an hour of the capture. The newest week can change until its Sunday has been captured.

Where does the data come from?

Each provider's own published pricing: pricing pages, documentation and pricing APIs, collected by GetDeploying and converted to US dollars at the European Central Bank's daily reference rate.

How is this different from the API?

The dataset is the weekly market aggregate for the trailing year, with no stability promise: its shape can change at any time. The API has per-provider prices, the daily interval, the full history and a documented contract.

See also the GPU price trends, our methodology and the GPU models dataset. Prices are estimates for information only, not an offer to sell compute.