GPU Cloud Price Index · snapshot 2026-08-06

What a GPU actually costs per hour, right now

32 on-demand prices verified 2026-08-06 across 3 providers (Lambda, RunPod, CoreWeave) — with the quote basis stated (per-GPU vs per-node), because mixing them is how every other comparison lies to you. Cheapest-first, every row sourced. Download CSV/JSON · CC BY 4.0

Quote basis: per GPU-hour

Per-GPU-hour (as published per-GPU by the provider)

GPUProvider$/GPU-hr Config noteSource · verified
NVIDIA RTX A5000RunPod$0.27Secure Cloud podsource · 2026-08-06
NVIDIA L4RunPod$0.39Secure Cloud podsource · 2026-08-06
NVIDIA A40RunPod$0.44Secure Cloud podsource · 2026-08-06
NVIDIA RTX 3090RunPod$0.46Secure Cloud podsource · 2026-08-06
NVIDIA RTX 4090RunPod$0.69Secure Cloud podsource · 2026-08-06
NVIDIA RTX 6000 AdaRunPod$0.77Secure Cloud podsource · 2026-08-06
NVIDIA Tesla V100Lambda$0.798× configsource · 2026-08-06
NVIDIA L40SRunPod$0.99Secure Cloud podsource · 2026-08-06
NVIDIA RTX 5090RunPod$0.99Secure Cloud podsource · 2026-08-06
NVIDIA A6000Lambda$1.091×–4× instancessource · 2026-08-06
NVIDIA A10Lambda$1.291× instancesource · 2026-08-06
NVIDIA A100 PCIeRunPod$1.39Secure Cloud podsource · 2026-08-06
NVIDIA A100 SXMRunPod$1.49Secure Cloud podsource · 2026-08-06
NVIDIA A100 40GBLambda$1.99PCIe/SXM configssource · 2026-08-06
NVIDIA RTX PRO 6000RunPod$1.99Secure Cloud podsource · 2026-08-06
NVIDIA GH200Lambda$2.291× instancesource · 2026-08-06
NVIDIA A100 SXM 80GBLambda$2.798× configsource · 2026-08-06
NVIDIA H100 PCIeRunPod$2.89Secure Cloud podsource · 2026-08-06
NVIDIA H100 SXMRunPod$2.99Secure Cloud podsource · 2026-08-06
NVIDIA H100 PCIeLambda$3.291× instancesource · 2026-08-06
NVIDIA H100 SXMLambda$4.291× instance; 8× config $3.99/GPUsource · 2026-08-06
NVIDIA H200RunPod$4.39Secure Cloud podsource · 2026-08-06
NVIDIA B200RunPod$5.89Secure Cloud podsource · 2026-08-06
NVIDIA B200 SXMLambda$6.991× instance; 8× config $6.69/GPUsource · 2026-08-06
NVIDIA B300RunPod$7.39Secure Cloud podsource · 2026-08-06
Quote basis: per instance-hour

Per-instance quotes (multi-GPU nodes — NOT normalized)

CoreWeave publishes node prices (HGX SKUs are typically 8-GPU). We deliberately don't divide them into per-GPU numbers — config details (interconnect, CPU, storage) make naive division misleading. Compare node-to-node.

InstanceProvider $/instance-hrNote Source · verified
NVIDIA GH200CoreWeave$6.50on-demandsource · 2026-08-06
NVIDIA L40S (node)CoreWeave$18.00on-demand; spot $7.88source · 2026-08-06
NVIDIA A100 (node)CoreWeave$21.60on-demand; spot $9.65source · 2026-08-06
NVIDIA GB200 NVL72 (node)CoreWeave$42.00on-demandsource · 2026-08-06
NVIDIA HGX H100 (node)CoreWeave$49.24on-demand; spot $19.71source · 2026-08-06
NVIDIA HGX H200 (node)CoreWeave$50.44on-demand; spot $20.93source · 2026-08-06
NVIDIA HGX B200 (node)CoreWeave$68.80on-demand; spot $34.11source · 2026-08-06

Workload cost calculator

Use this data — 32 verified prices, CC BY 4.0

The GPU Cloud Price Index snapshot of 2026-08-06: 25 per-GPU-hour rows and 7 per-instance rows across 3 providers, each with its config note, source URL and verification date.

License: CC BY 4.0 — reuse, republish or build on this table anywhere (commercial use included); attribute “AnswerFootprint GPU Cloud Price Index” and link https://answerfootprint.com/gpu/.

AnswerFootprint. “GPU Cloud Price Index” (snapshot 2026-08-06). https://answerfootprint.com/gpu/ — CC BY 4.0.

Methodology (short, because it's simple)

  • Verified-rows-only: every price re-fetched from the provider's public pricing page and date-stamped. Providers whose pricing isn't publicly fetchable (JS-only consoles, contact-sales SKUs) are excluded, not estimated.
  • Quote basis never mixes: per-GPU and per-instance tables stay separate; node prices are never divided into per-GPU numbers.
  • On-demand only in v1; spot/reserved noted where the provider publishes them.

Prices belong to providers and change often; every row above is date-stamped and re-verified on refresh. Referral links (when live) never reorder these tables. Stale price? Tell us — it outranks everything.

Frequently asked questions

What is the difference between per-GPU and per-instance pricing?

A per-GPU-hour price is what the provider publishes for one GPU. A per-instance price covers a whole multi-GPU node — CoreWeave's HGX SKUs are typically 8 GPUs plus interconnect, CPU and storage. We keep the two in separate tables and never divide a node price into a per-GPU number, because config differences make naive division misleading.

Why does the same GPU cost different amounts at different providers?

Because the SKUs are not identical even when the chip is. A verified H100 SXM ranges from $2.99/GPU-hr (RunPod Secure Cloud pod) to $4.29 (Lambda 1× instance) in this index — the spread comes from what surrounds the GPU: instance size (1× vs 8×), PCIe vs SXM, included CPU/storage, and the provider's platform. The config-note column states each row's basis so you compare like with like.

What does “verified” mean on this page?

Every price was re-fetched from the provider's public pricing page on 2026-08-06 and date-stamped; each row links the exact source page. Providers whose pricing is not publicly fetchable (JS-only consoles, contact-sales SKUs) are excluded rather than estimated.

Can I reuse this data?

Yes. The full index — 32 rows across 3 providers — is downloadable as CSV or JSON and licensed CC BY 4.0: reuse it anywhere, with attribution to AnswerFootprint and a link to this page.