Aethir review

Entry is $1.25 for a GPU hour, billed hourly. Work is counted in hourly steps rather than by the second, so steady long running jobs fit the model better than quick experiments. You only add your own transfer costs for data you move elsewhere.

The published lineup in B200, H200 and H100 spans the top datacenter accelerators plus L40S for lighter inference work. The part you actually get depends on availability in the region you deploy to, so leave some slack in your model choice.

The published position is no hidden bandwidth fees, so data leaving the platform does not turn into a separate bandwidth line. That matters most for training jobs that pull large datasets and for inference servers that return big responses.

The stated model is bare metal, no virtualization overhead, which removes a virtualization layer between your job and the accelerator. For training runs and latency sensitive inference that usually means steadier throughput, and you can size the work directly against the card you rent.

Support runs on email, help centre and community. That means a help centre and a community channel sit alongside email, so documentation and peer answers come first, and anything urgent on a paid training run goes out by email.

The terms name DCI Foundation as the contracting entity, with a headquarters in Panama. If you buy as a company, check that entity against your contract and invoice, since that is the party your legal team signs with.

Entry sits at $1.25, above the €0.61 median across the ranked GPU providers on the page. The gap buys bare metal hardware and bandwidth already in the rate, not the lowest headline number.

The rate card is hourly at $1.25 per hour, so a short paid trial is the practical way to size a job before a long schedule. Longer runs accumulate the same hourly rate for as long as the card is busy.