CoreWeave
GPU pricing, tracked daily since 2026-09-18On chips where at least three providers publish rates (8 of CoreWeave's 8 models), CoreWeave is at or below the panel median on 0 and holds the single cheapest published rate on 0.
CoreWeave vs RunPod · CoreWeave vs SpotGPUs · CoreWeave vs GPUBrazil · CoreWeave vs Verda (formerly DataCrunch) · CoreWeave vs Hugging Face · CoreWeave vs gpu.ai · CoreWeave vs Hostnot GPU · CoreWeave vs Spheron
Published per-instance on-demand $/hour rates in plain HTML on the pricing page's rate card. GPU rows are 8-GPU HGX nodes priced per node; the page's own 'Inference Single' price confirms the per-GPU rate is the node total divided by 8 (H100 $49.24/8 = $6.16). Rows are emitted at the stated node price and count. Excluded: GB200 NVL72 (rack-scale unit, per-GPU vs per-node ambiguous), B300 and RTX PRO 6000 Standard Memory (on-demand is contact-sales), CPU/storage/network rows, and all spot prices. North America and Europe tables carry identical on-demand rates, so one row per model with 'North America / Europe'. The page publishes rates, not stock - availability shows as catalog.