The cheapest H100 hour is $0.31 spot — but stable on-demand anchors sit at $2.46–$2.95
Asked:
“What is the real effective price per hour for an NVIDIA H100 across GPU cloud providers? Include the sticker price, spot or committed discounts, egress fees, and current availability for each provider.”
Ten providers with publicly verifiable H100 pricing, captured 2026-09-04, all in USD per GPU-hour (multi-GPU node prices divided by GPU count). Scaleway prices only in euros (€2.88/GPU-hour) and is excluded from the USD ranking. Azure had no stable public H100 price recoverable, so it has no row.
sticker / on-demand
committed / reserved
spot / preemptible / marketplace — interruptible
bar spans sticker → lowest price, sorted by lowest USD compute price
The headline "cheapest" rates are volatile: Verda/DataCrunch spot at $0.31 and Vast.ai marketplace at about $0.98 can be interrupted or vanish; treat CoreWeave $2.46, Lambda $2.59, Runpod $2.89 and Nebius $2.95 as the stable on-demand anchors.
api.datacrunch.io ·
coreweave.comOnly Runpod publishes an explicit egress rate in the captured pages — $0/GB — so its $1.99 Community Cloud rate holds even under an illustrative 10 GB/GPU-hour egress workload; every other egress cell is unverified, not zero.
runpod.ioHyperscalers price highest: Google's a3-highgpu-8g is $11.06 on demand per GPU, $6.64 spot, $4.86 with a 3-year commitment; AWS London Capacity Blocks are $4.72 and require reserving a fixed block.
cloud.google.com ·
aws.amazon.comAvailability is a public signal, not a live guarantee: CoreWeave is subject to capacity, Google is region/quota dependent, Vast.ai and Verda are dynamic marketplace inventory, and Lambda and Hyperstack expose no live inventory at all.
scaleway.com marks its €2.88 H100-SXM GA.
All captured rows
| Provider · product | Sticker $/h | Spot / market | Committed | Lowest | Egress $/GB | Effective @10 GB/h | Availability signal | Source |
|---|
Provider comparison of NVIDIA H100 GPU cloud pricing, 10 rows, captured 2026-09-04 from public pricing pages and APIs. All prices USD per GPU-hour; 8-GPU node prices divided by 8. Effective price = compute + egress GB/GPU-hour × egress $/GB, shown only where egress was verified (Runpod). Scaleway (€2.88/GPU-hour) excluded from the USD chart; Azure omitted for lack of a recoverable public price; storage, networking services, taxes and regional variation not captured.