Why H200 prices differ between providers

On 12 Nov 2026, 15 providers listed an on-demand H200 price. The lowest was $3.49 per GPU-hour and the highest $10.60, with a median of $5.40.

The same card sits behind every one of those figures. What differs is everything else.

A provider that rents single cards and one that rents only servers of eight both print a per-GPU price. The first is a bill. The second is one eighth of a bill. Large machines often show a lower per-GPU figure, and need a larger commitment of money each hour.

The GPU is one part of the machine. CPU cores, system memory, local disk and network speed vary widely at the same GPU price. So does what the provider adds: images ready to use, storage that persists, support that answers.

A lower price can mean a thinner machine. Read the instance details, not only the headline.

Some price pages lead with the reserved rate, because it is the lowest number available. Set beside another provider’s on-demand rate, it looks like a large saving. It is a different product. The tables on this site list on-demand prices only.

A price is cheap to publish and costs nothing to keep online when the machines are gone. So some of the lowest figures belong to listings that cannot be started. The labels on this site, Start now, No GPUs, Gated, Quota or Not checked, separate the two.

Fix the number of GPUs you need. Drop every listing whose smallest rental is larger. Drop every listing you cannot start. Then compare hourly totals in the unit each one bills in. The few rows left are your real choices.

Runcrate lists each GPU with its own price per GPU-hour, read from its live feed when a page is built. Billing is by the second on most GPUs, and booting is never billed. Runcrate can start an H200 now from $4.67 per GPU-hour.

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.