Independent reference. We are independent of every vendor listed. No affiliate links. No sponsored placements.
Methodology - Last verified June 2026

How we source GPU cloud pricing data

Every rate on this site is sourced from a vendor's public pricing page or its public cloud retail price list. Source URLs are listed below. Re-verification cadence is monthly for the top-traffic vendors and quarterly for the long tail. The verified date appears in every page footer.

Normalisation

Where vendors publish per-node or per-instance rates (typical at hyperscalers, where p5.48xlarge is an 8x H100 SXM node and is billed as a single unit), we divide by the number of GPUs to produce a per-GPU per-hour rate. This is the apples-to-apples comparison buyers want. Per-GPU normalisation does not represent the smallest billable unit at the hyperscalers (you cannot buy 1 GPU out of an 8-GPU node). We note this in the configuration column on each comparison table.

Quote-only vendors

For vendors that do not publish per-GPU per-hour rates (Fly.io GPU, OVHcloud GPU as of June 2026), we mark the vendor as quote-only, label every rate field as "Quote only - contact [vendor] sales," and link to the vendor contact page. We do not infer rates.

Industry benchmarks

Where we cite a market-wide claim (analyst rankings, market share, etc.) we name the published dataset (Gartner Magic Quadrant, IDC MarketScape, Forrester Wave, BLS OEWS, Damodaran NYU Stern, etc.) and the report year. We do not use phrases like "industry average" or "typical based on observation" without a citation.

Source list (17 vendors)

Last verified June 2026.