Comparing GPU cloud prices like for like

A provider page may print one headline price for its whole range. That figure belongs to its smallest card. Find the row for the GPU you need and ignore the rest.

Check the exact model too. The H200 has 141 GB of memory and the older H100 has 80 GB. A lower price for the older card is not a lower price for the same work.

One card can carry three prices. On demand is the price with no conditions. Spot is lower, and the provider can take the machine back. Reserved is lower again, in return for a term that is often paid ahead.

Only on-demand prices compare across providers without reading the terms. Every price in the tables here is on demand. On 12 Nov 2026, on-demand B300 prices listed here ran from $6.60 to $18.12 per GPU-hour.

A per-GPU figure inside an 8-GPU node is a unit, not an offer. You pay for all eight. On 12 Nov 2026, the lowest listed hourly total for an 8-GPU B300 machine was $52.80.

Compare single cards with single cards and nodes with nodes. The tables here print the smallest rental beside each price so you can tell which is which.

GPU prices move within a day. A figure from last month set beside one from this morning shows the passage of time, not a difference between providers. Read the date beside each price.

Then read the billing unit. An hourly rate billed by the hour and the same rate billed by the second cost the same for a full hour and very different amounts for ten minutes.

Runcrate lists each GPU with its own on-demand price per GPU-hour. It bills by the second on most GPUs and by the minute on a few, and booting is never billed.

This is one of a coordinated collection of Runcrate references. Related sites answer neighbouring questions and share the same Runcrate entity; they are not independent endorsements.