AIInfraDecider

Methodology

How we research and compare

Our comparisons are built from what providers publish: pricing pages, instance catalogues, documentation and region pages. We do not rent GPUs to benchmark them, and no page claims otherwise.

The GPU Cloud Index

Each row of the index is one provider, one GPU and one way of renting it (on-demand, spot, community, marketplace, reserved or serverless). For every row we record:

Inclusion depends on public information only: a provider is listed when it publishes at least one GPU offering. Inclusion is never paid for.

What we never do

Normalisation

Per-second or per-minute prices are converted to per-hour. Prices for multi-GPU instances are divided by the number of GPUs. Marketplace prices are set by individual hosts; we show the lowest listing visible on the verification date and mark it. Prices exclude tax, and storage and network charges unless a row says otherwise.

Updates

We update a row when its source changes: a price change, a new GPU, a region launch or exit, a change to the billing model, or a provider launch, rename, acquisition or shutdown. A check that finds no change does not change the page's public date.

Tools

Research and drafting are tool-assisted. The editorial desk checks every published figure against its source before publication.

Money

We may earn affiliate commissions and accept clearly labelled sponsorship. Nobody can buy a ranking, a "choose this" recommendation, a price figure, an index position or the removal of a competitor. See the disclosure policy and partner policy.

Corrections

If a figure is wrong or out of date, use the contact form with the page and the source. Corrections are free.