Skip to main content
GET
GPU Prices (Historical)

Overview

The GPU Prices API lets you pull the daily rental price of one GPU model at a time: what an hour of H100, H200, B200, or B300 compute costs on the open market, in US dollars per GPU hour. Each day carries the spread across providers behind the price and the change over 1, 7, and 30 days, computed for you. The index is a reference rate built from the published on-demand prices of many cloud GPU providers. A new value is published every 15 minutes; we store one value per day, the newest print of each UTC day, and refresh today’s value every hour. To get started, please create an account and grab your API key at financialdatasets.ai. You will use the API key to authenticate your API requests.

Coverage

Series

You can fetch this list, with each series’ first available date, with a free GET request to: https://api.financialdatasets.ai/macro/gpu-prices/series/

Reading the fields

  • date is the UTC day the value belongs to. The value is the newest print of that day; for today it is the newest print at the last hourly refresh, and as_of tells you when that was.
  • value is the price in US dollars per GPU hour.
  • stability_band is the standard deviation of the providers’ prices behind value, in the same units. A wide band means providers disagree; a narrow band means a tight market.
  • change_1d is value minus the prior day’s value, in dollars. change_1d_pct, change_7d_pct, and change_30d_pct are the percent changes from 1, 7, and 30 days earlier, rounded to two decimals.
  • A change is null when the comparison day has no value.
  • attribution names the source of the index. Please keep it when you display the data.

Source and license

The index is the Computable GPU Index, published by Computable under a Creative Commons Attribution license. Attribution: “Computable GPU Index by Computable — mcp.getcomputable.com”.

Getting Started

There are only 2 steps for making a successful API call:
  1. Add your API key to the header of the request as X-API-KEY.
  2. Execute the API request with series and, optionally, start_date and end_date.

Example

GPU Prices

Example Response

Rows come newest first, ten per page; follow next_page_url for the rest of the window. Without dates the window is the trailing 90 days.

Authorizations

X-API-KEY
string
header
required

API key for authentication.

Query Parameters

series
string
required

The GPU model: h100, h200, b200, or b300. Use the /macro/gpu-prices/series endpoint to get the list.

start_date
string

The first day to return, inclusive, in YYYY-MM-DD format. Defaults to 90 days before end_date when both dates are omitted; omit it with an end_date to get all available history through that day.

end_date
string

The last day to return, inclusive, in YYYY-MM-DD format. Defaults to today.

cursor
string

Opaque pagination cursor from a previous response's next_page_url. When provided, all other query parameters are ignored: the cursor carries the original request's filters. Treat it as opaque; do not construct or modify it.

Response

GPU prices response

gpu_prices
object[]
attribution
string

The source credit to keep when displaying the data: the Computable GPU Index.

next_page_url
string

Absolute URL of the next page of results. Present only when more results remain; request it as-is to continue. Each page holds up to 10 records.