How GPUHunt Works
Our data sources, collection methodology, and how we normalize pricing across 20+ providers.
What we track
GPUHunt aggregates GPU and bare metal server pricing from 20+ cloud providers including Lambda Labs, CoreWeave, RunPod, Vast.ai, DigitalOcean, Hetzner, Scaleway, Oblivus, and more. We focus exclusively on GPU compute relevant to AI/ML workloads — H100, A100, MI300X, L40S, RTX 4090, and similar chips used for training, fine-tuning, and inference.
How pricing is collected
Prices are scraped directly from provider pricing pages and APIs on an automated schedule. For providers with public APIs, we query current on-demand pricing directly. For providers without APIs, we maintain curated price tables that are manually verified and updated regularly. All prices shown are on-demand (pay-as-you-go) unless otherwise noted. Reserved, spot, or contract pricing is not shown unless explicitly labeled.
Price normalization
All prices are normalized to a per-hour basis. Monthly prices are calculated as hourly × 730 (average hours/month). Prices in EUR are stored and displayed with the € symbol; USD prices use $. We do not perform live currency conversion — EUR-priced providers (e.g. Hetzner, OVHcloud, Scaleway) show their native pricing. GPU model names are normalized to a canonical form (e.g. "NVIDIA H100 SXM5", "AMD Instinct MI300X") to enable cross-provider comparison.
$/GB VRAM metric
The $/GB·hr column shows the hourly cost per gigabyte of total GPU VRAM in the configuration. For multi-GPU nodes, total VRAM = per-GPU VRAM × GPU count. This metric lets engineers compare the true memory cost across different GPU types and configurations. An H100 80GB at $2.50/hr costs $0.031/GB·hr, while an RTX 4090 24GB at $0.40/hr costs $0.017/GB·hr — useful when your bottleneck is VRAM, not compute.
Availability
Availability status reflects what we observed at last scrape time. "Available" means the provider listed the configuration as orderable at time of scrape. Actual availability may differ — some providers show all SKUs regardless of stock. We recommend checking directly with the provider before planning workloads around specific configurations.
What we don't show
We do not include: spot/preemptible pricing (too variable), custom enterprise contracts, academic or startup credit programs, or bundled pricing that includes storage/networking separately from compute. We do not show CPU-only instances. Vast.ai and Salad Cloud are marketplace providers where individual users list their own hardware — their pricing is shown in a separate section and may not reflect the same reliability guarantees as cloud providers.
Affiliate disclosure
Some provider links on GPUHunt include affiliate tracking parameters. When you click through and sign up or make a purchase, we may receive a referral commission from the provider. This does not affect pricing, ranking, or how providers are ordered in results — all comparison tables are sorted by price, not affiliate relationship. We disclose all affiliate relationships.
Errors and corrections
If you find a pricing error, missing provider, or incorrect GPU spec, please email support@gpu-hunt.com. We take data accuracy seriously and will correct verified errors within 24 hours. Provider pricing can change without notice — always verify current pricing directly with the provider before committing to a workload.
Pricing data changes frequently. If something looks wrong, let us know and we'll fix it.
Report an issue