GPUPriceBook

Cheapest H100 rental per hour

The cheapest on-demand NVIDIA H100 rental in June 2026 is $1.90/hr (H100 PCIe) on Hyperstack. On spot/community tiers it falls to about $0.95/hr on Vast.ai. Specialist neoclouds and the Vast.ai marketplace dominate the low end; AWS, GCP and Azure sit far higher because they sell H100s as whole 8-GPU instances bundling vCPU/RAM/support. See the full ranking below.

Source: Provider pricing pages. Data as of June 2026.

H100 prices ranked cheapest first

RankProviderVariantOn-demand /hrSpot/community /hr
#1HyperstackH100 PCIe$1.90/hr$1.52/hr
#2Vast.aiH100 SXM$1.93/hr$0.95/hr
#3Vultr Cloud GPUH100 SXM$2.30/hr
#4RunPodH100 PCIe$2.39/hr$1.99/hr
#5HyperstackH100 SXM$2.40/hr
#6Vast.aiH100 PCIe$2.78/hr$1.40/hr
#7Verda (formerly DataCrunch)H100 SXM$3.25/hr
#8LambdaH100 SXM$3.29/hr
#9RunPodH100 SXM$3.29/hr$1.99/hr
#10Paperspace (DigitalOcean)H100 SXM$3.39/hr
#11NebiusH100 SXM$3.85/hr$2.15/hr
#12CrusoeH100 SXM$3.90/hr
#13Together AIH100 SXM$4.79/hr
#14CoreWeaveH100 SXM$6.16/hr
#15AWS EC2H100 SXM$6.88/hr$3.37/hr
#16Microsoft AzureH100 PCIe$6.98/hr$1.75/hr
#17Google CloudH100 SXM$10.98/hr$4.98/hr
#18Microsoft AzureH100 SXM$12.29/hr$2.27/hr

Source: Provider pricing pages. Data as of June 2026.

Snapshot June 2026 — cloud GPU prices change weekly; verify on the provider's pricing page before relying on a figure.

Frequently asked questions

What is the cheapest H100 rental per hour in 2026?

The cheapest on-demand H100 we track is $1.90/hr (H100 PCIe) on Hyperstack. On spot/community tiers it drops to about $0.95/hr on Vast.ai. Snapshot June 2026 — cloud GPU prices change weekly; verify on the provider's pricing page before relying on a figure.

Why is the H100 so much cheaper on neoclouds than AWS?

Specialist neoclouds (RunPod, Vast.ai, Hyperstack, Vultr, Lambda) sell GPU compute with minimal bundled services, so an H100 runs ~$2-4/hr. Hyperscalers price flagship GPUs only as whole 8-GPU instances bundling lots of vCPU, RAM and support, pushing the per-GPU figure to $7-12/hr.

SXM vs PCIe H100: which should I rent?

SXM has higher memory bandwidth and NVLink for multi-GPU scale-out, so it is preferred for large distributed training. PCIe (350W) is fine for single-GPU training and inference and is sometimes cheaper. Most neoclouds list SXM; only a few (RunPod, Hyperstack, Vast.ai) publish PCIe.

Is spot H100 worth it?

If your job checkpoints and can survive interruption, spot/community H100 (often ~50% off on-demand) is great value. For long unattended runs where a restart is costly, on-demand or reserved is safer.

Keep exploring

Last updated: 2026-06-21