Provider coverage and gaps
This pass contains 351 configurations across 22 providers. Coverage follows exact public SKUs and presets. The table below shows where current USD rates and deployment regions were captured; catalogue coverage does not establish live stock.
Sources accessed . Read the verification method.
Scope decisions
Rows include full-GPU VMs, provider presets, bare-metal shapes and dedicated servers. Published CPU/RAM variants are distinct configurations. Together rows represent a whole GPU node; RunPod rows use the one-GPU Secure Cloud Pod tariff. A model cell remains empty when a provider publishes GPU memory and count without identifying the device.
Hetzner publishes EUR hourly rates with setup fees and monthly caps. Scaleway’s captured L4 rates are also in EUR. These stay in native-currency fields; USD cells remain empty without a sourced conversion. Fractional GPU slices, inference APIs, regional duplicates and contract-only offers are outside this pass.
Partial coverage and exclusions
Vast.ai. Host-set marketplace offers vary by machine, GPU count, CPU share, location and rental terms. The public price cards are market floors, not fixed instance configurations. No stable offer snapshot is included. Source checked 2026-11-11.
Vultr. The public API fully identified the L40S plan’s model, physical GPU count and memory. Other GPU plan identifiers omit physical count; H100/B200 catalogue entries also reported deploy_ondemand=false. Those entries are excluded rather than decoded into assumed hardware or current rates. The L40S locations array was empty. Source checked 2026-11-11.
Genesis Cloud. Published RTX 3090 and H100 presets are included. The H100 heading establishes its device memory; RTX 3090 device memory remains empty because the instance table only specifies host RAM. Current public tariffs were not verified. Source checked 2026-11-11.
Fluidstack. No fixed public instance configuration with enough sourced specification fields was established. Enterprise clusters and negotiated capacity are outside this instance catalogue. Source checked 2026-11-11.
AWS Trainium / Inferentia. These are custom AI accelerators rather than GPUs. They are outside this GPU catalogue; compare their accelerator architecture separately before comparing purchase prices. Source checked 2026-11-11.
Continue the buying research
Neocloud Companies compares the operators behind GPU capacity. HPC Providers covers managed clusters, schedulers and services. Custom Silicon List covers accelerator architectures beyond GPUs.
Return to the instance table or download CSV and JSON with field-level sources.