gpu.gl

NVIDIA H100 SXM cloud prices

H100 SXM on-demand cloud prices run from $4.41 to $10.00 per GPU-hour across 5 providers gpu.gl tracks (median $4.94). The lowest is DigitalOcean.

On-demand$4.41low · median $4.94 · high $10.00 · 5 providers
Spot / interruptible$2.47low · median $2.47 · high $2.47 · 1 provider
Reserved 1 yr$3.26low · median $3.26 · high $3.26 · 1 provider
Reserved 3 yr$2.97low · median $2.97 · high $2.97 · 1 provider

Lowest price by provider

H100 SXM prices by provider
ConfigurationWhereFetchedSource
DigitalOceanOn-demand $4.41 NVIDIA HGX H100 GPU Droplet · 1 GPU Not stated per region DigitalOcean ↗
NebiusOn-demand $4.50 HGX H100 · 1 GPU Not stated per region Nebius ↗
DataCrunchOn-demand $4.94 4H100.80S.176V · 1, 2, 4, 8 GPUs Not stated per region DataCrunch ↗
AWSOn-demand $6.88 p5.48xlarge · 1, 8 GPUs 5 regions AWS ↗
OCIOn-demand $10.00 B98415 (Compute - GPU - H100) · 1 GPU All commercial regions OCI ↗
DataCrunchSpot / interruptible $2.47 8H100.80S.176V · 1, 2, 4, 8 GPUs Not stated per region DataCrunch ↗
DigitalOceanReserved 1 yr $3.26 NVIDIA HGX H100 GPU Droplet · 1 GPU Not stated per region DigitalOcean ↗
AWSReserved 3 yr $2.97 p5.48xlarge · 1, 8 GPUs 4 regions AWS ↗

Price history

History starts 2026-11-08. gpu.gl stores one snapshot a day from launch and never back-fills, so a chart appears here once there are two daily snapshots with on-demand prices for this GPU (1 so far).

H100 SXM specifications

The H100 SXM has 80 GB of HBM3 and 3.35 TB/s of memory bandwidth. That is about 295 dense FP16 TFLOPS per TB/s of bandwidth; workloads that stream large weights, like LLM inference at small batch sizes, tend to be limited by the bandwidth figure, while large-batch training leans on the compute figure.

ArchitectureHopper
Memory80 GB HBM3
Memory bandwidth3.35 TB/s
FP16 / BF16 tensor (dense)989.5 TFLOPS
FP8 tensor (dense)1,979 TFLOPS
Max power700 W
Form factorSXM5

Source: NVIDIA spec sheet. Tensor figures are dense (without sparsity).

Every configuration

Each instance size or listing, at its cheapest region.

ProviderPrice type$/GPU-hrInstanceGPUsInstance priceCheapest regionFetchedSource
DigitalOceanOn-demand$4.41NVIDIA HGX H100 GPU Droplet1$4.41 per GPU-hourNot stated per regionDigitalOcean ↗
NebiusOn-demand$4.50HGX H1001$4.50 per GPU-hourNot stated per regionNebius ↗
DataCrunchOn-demand$4.944H100.80S.176V4$19.75/hrNot stated per regionDataCrunch ↗
DataCrunchOn-demand$4.948H100.80S.176V8$39.50/hrNot stated per regionDataCrunch ↗
DataCrunchOn-demand$4.941H100.80S.30V1$4.94/hrNot stated per regionDataCrunch ↗
DataCrunchOn-demand$4.941H100.80S.32V1$4.94/hrNot stated per regionDataCrunch ↗
DataCrunchOn-demand$4.942H100.80S.80V2$9.88/hrNot stated per regionDataCrunch ↗
AWSOn-demand$6.88p5.48xlarge8$55.04/hrUS East (N. Virginia) +4 moreAWS ↗
AWSOn-demand$6.88p5.4xlarge1$6.88/hrUS East (N. Virginia) +4 moreAWS ↗
OCIOn-demand$10.00B98415 (Compute - GPU - H100)1$10.00 per GPU-hourAll commercial regionsOCI ↗
DataCrunchSpot / interruptible$2.478H100.80S.176V8$19.75/hrNot stated per regionDataCrunch ↗
DataCrunchSpot / interruptible$2.471H100.80S.30V1$2.47/hrNot stated per regionDataCrunch ↗
DataCrunchSpot / interruptible$2.471H100.80S.32V1$2.47/hrNot stated per regionDataCrunch ↗
DataCrunchSpot / interruptible$2.472H100.80S.80V2$4.94/hrNot stated per regionDataCrunch ↗
DataCrunchSpot / interruptible$2.474H100.80S.176V4$9.88/hrNot stated per regionDataCrunch ↗
DigitalOceanReserved 1 yr$3.26NVIDIA HGX H100 GPU Droplet1$3.26 per GPU-hourNot stated per regionDigitalOcean ↗
AWSReserved 3 yr$2.97p5.48xlarge8$23.78/hrUS East (N. Virginia) +3 moreAWS ↗
AWSReserved 3 yr$2.97p5.4xlarge1$2.97/hrUS East (N. Virginia) +3 moreAWS ↗

Compare