Vast.ai review

Median entry pricing across the marketplace sits at €0.63 per GPU hour, with €0.39 at the lower end and €1.75 at the upper. Listings on the site start as low as €0.01, so the card you pick moves the number more than the platform does.

Charges accrue per second, so an instance that runs for a short burst is not billed for a whole hour. Across GPU listings on the site, 73.6% and 74.5% are the most common billing shapes, with 35.5% also in use.

The catalogue includes RTX 3060, RTX 4090 and A100, so consumer RTX parts and data-centre silicon sit side by side. Most GPU listings on the site run Linux, and 70.6% of them also cover Windows. Prices differ by host, so the same model can carry a different rate from one machine to the next.

The egress fee is set by each host as a per-byte charge for data sent and received. Data-heavy jobs should read the fee on the machine they pick, because the rate belongs to the host and changes with the machine you rent.

Capacity is spread across from 40 locations, with a marketplace median of 4 and an upper quartile of 10. Where a job runs follows the host you choose, so latency and local data rules come with the machine rather than a fixed region list.

Support runs on email, chat and help centre, so email, chat, a help centre and community answers are the routes in. Across GPU listings on the site, 76.9% and 78.9% are the most common channels. Phone is not one of the listed routes.

Vast.ai publishes SOC 2 Type 2, SOC 3 and HIPAA as its compliance certifications. The marketplace runs as Vast.ai Inc., headquartered in the United States, and was founded in 2018. Those certifications describe the operator, so teams with vendor questionnaires can weigh them against their own requirements.

A single vendor such as SberCloud lists B300, H100 and A100 under one roof with its own pricing. Vast.ai is a marketplace, so the card, its rate and its egress charge come from the host. Choose it when price flexibility matters more than a fixed catalogue.