17 providers · per GPU-hour
NVIDIA H100 cloud pricing
19 offers from 17 providers, 18 with a published on-demand price. Per GPU-hour, US$. Includes H100 (SXM, HGX) and H100 NVL. Prices checked 28 Sept 2026.
- Lowest on-demand
- $1.99/GPU-hr
- Lowest spot
- $0.74/GPU-hr
- Memory
- 80 GB HBM3
Spec sheet: H100 (SXM, HGX) ↗H100 NVL ↗
| # | Provider | Offer | On-demand | Spot | Committed | As published (original currency) · source | Actions |
|---|---|---|---|---|---|---|---|
| 1 | Voltage Park | H100 x1 1× H100 | $1.99/GPU-hr | — | — | ||
| 2 | UpCloud | H100 x1 1× H100 · 12 vCPU · 240 GB RAM | $2.00/GPU-hr | $1.99/GPU-hr | — | ||
| 3 | Krutrim Cloud | H100 x1 1× H100 · 24 vCPU · 200 GB RAM | $2.25/GPU-hr | — | $1.82/GPU-hr 1-yr reserved | ||
| 4 | Hyperstack | H100 x1 1× H100 | $2.50/GPU-hr | $2.00/GPU-hr | — | ||
| 5 | AceCloud | H100 x1 1× H100 · 26 vCPU · 250 GB RAM | $2.60/GPU-hr | — | — | ||
| 6 | E2E Networks | H100 x1 1× H100 · 26 vCPU · 250 GB RAM | $2.69/GPU-hr | $0.74/GPU-hr | $1.64/GPU-hr Committed (3-mo) | ||
| 7 | Jarvislabs | H100 x1 1× H100 · 16 vCPU · 200 GB RAM | $2.69/GPU-hr | — | — | ||
| 8 | Scaleway | H100 x1 1× H100 · 24 vCPU · 240 GB RAM | $3.06/GPU-hr | — | — | ||
| 9 | Verda (formerly DataCrunch) | H100 x1 1× H100 | $3.27/GPU-hr | $1.63/GPU-hr | — | ||
| 10 | Yotta Infrastructure / Shakti Cloud | VM – 4 x H100 On Demand 4× H100 · 102 vCPU · 960 GB RAM · 100 GB disk | $3.75/GPU-hr | — | — | INR 1,423/hr | |
| 11 | Yotta Infrastructure / Shakti Cloud | VM – 1 x H100 On Demand 1× H100 · 24 vCPU · 240 GB RAM · 100 GB disk | $3.75/GPU-hr | — | — | INR 356/hr | |
| 12 | Yotta Infrastructure / Shakti Cloud | VM – 2 x H100 On Demand 2× H100 · 52 vCPU · 480 GB RAM · 100 GB disk | $3.75/GPU-hr | — | — | INR 712/hr | |
| 13 | Nebius | H100 x1 1× H100 | $3.85/GPU-hr | $2.15/GPU-hr | — | ||
| 14 | Lambda | H100 H100 · 208 vCPU | $3.99/GPU-hr | — | — | ||
| 15 | Neysa | H100 x1 1× H100 · 48 vCPU · 256 GB RAM · 1000 GB disk | $4.39/GPU-hr | — | $2.44/GPU-hr 3-yr reserved | ||
| 16 | RunPod | H100 serverless 1× H100 | $4.79/GPU-hr | — | — | ||
| 17 | CoreWeave | NVIDIA HGX H100 (8x) 8× H100 · 128 vCPU · 2048 GB RAM | $6.16/GPU-hr | $2.46/GPU-hr | — | ||
| 18 | Oracle Cloud Infrastructure | H100 x1 1× H100 | $10.07/GPU-hr | — | — | ||
| 19 | Vultr | H100 x8 8× H100 | Contact for pricing | — | $1.99/GPU-hr Committed (2-yr) |
What is the H100 used for?
The H100 is a data-centre GPU positioned by its maker for large-scale model training, built on the Hopper architecture and released in 2022. Multi-GPU servers link cards over NVLink 4th gen (900 GB/s), which matters when a job is split across several GPUs.
How large a model fits on one H100?
With 80–94 GB of memory, one card holds the weights of a model up to about 40–47 billion parameters at 16-bit precision, or about 80–94 billion at 8-bit. That is weights only: inference also needs room for the KV cache, and training needs several times more for gradients and optimiser state, so larger jobs are split across multiple GPUs.
| Spec (as published) | H100 (SXM, HGX) | H100 NVL |
|---|---|---|
| Architecture | Hopper | Hopper |
| Form factor | SXM | PCIe dual-slot, air-cooled |
| Memory | 80 GB HBM3 | 94 GB HBM3 |
| Memory bandwidth | 3,350 GB/s | 3,900 GB/s |
| Interconnect | NVLink 4th gen (900 GB/s) | NVLink bridge (600 GB/s) |
| Max power (TDP) | 700 W | 400 W |
| Precisions listed | FP64, TF32, FP8, INT8 | — |
| Partitioning | Up to 7 MIGs @ 10GB | Up to 7 MIGs @ 12GB |
| Released | 2022 | 2023 |
| Source | maker spec sheet ↗ | maker spec sheet ↗ |
Need H100 capacity for a longer run?
Committed and cluster pricing is often negotiated. Tell us the GPU count, duration and region.