Data-center GPUs are not sold in shops; their market price is the hourly rate the cloud providers list. Here is that rate for every GPU we can price, from each provider's own price list.
Microsoft Azure, Oracle Cloud, Google Cloud · 27 instance types · last read Nov 3, 13:01 UTC
No list-price changes since we began reading the price lists on Nov 3, 2026. When a provider moves a price, the old and new rate appear here with the time we read it.
Every instance type we read
A cloud rents GPUs in instances of one, two, four or eight. The per-GPU-hour price is the instance price divided by its GPUs.
Every GPU instance type read, with its on-demand price
Microsoft Azure Retail Prices API: Linux virtual machines on demand, the cheapest US region for each instance type.
Oracle Cloud Public price list: GPU compute priced per GPU-hour.
Google Cloud Cloud Billing Catalog: Compute Engine GPU prices on demand in us-central1 (Iowa), per GPU.
Prices are list prices in US dollars before tax, read twice a day. Committed-use, reserved and spot prices are lower and are not shown; a price not read for a week is dropped.
Each GPU links to its page in the hardware tracker, which follows the maker's own listing.
On-demand list price per GPU-hour from each provider's official price list, read Nov 3, 13:01 UTC. Instance prices are divided by their GPUs; Azure is Linux in its cheapest US region. Committed-use and spot prices are lower.