8 providers · per GPU-hour
NVIDIA H200 cloud pricing
8 offers from 8 providers, 8 with a published on-demand price. Per GPU-hour, US$. Includes H200 (SXM, HGX) and H200 NVL. Prices checked 28 Sept 2026.
- Lowest on-demand
- $3.22/GPU-hr
- Lowest spot
- $2.45/GPU-hr
- Memory
- 141 GB HBM3e
Spec sheet: H200 (SXM, HGX) ↗H200 NVL ↗
| # | Provider | Offer | On-demand | Spot | Committed | As published (original currency) · source | Actions |
|---|---|---|---|---|---|---|---|
| 1 | AceCloud | H200 x1 1× H200 · 16 vCPU · 128 GB RAM | $3.22/GPU-hr | — | — | ||
| 2 | E2E Networks | H200 x1 1× H200 · 30 vCPU · 375 GB RAM | $3.99/GPU-hr | — | — | ||
| 3 | Hyperstack | H200 x1 1× H200 | $3.99/GPU-hr | — | — | ||
| 4 | Jarvislabs | H200 x1 1× H200 · 28 vCPU · 300 GB RAM | $3.99/GPU-hr | — | — | ||
| 5 | Verda (formerly DataCrunch) | H200 x1 1× H200 | $4.24/GPU-hr | — | — | ||
| 6 | Nebius | H200 x1 1× H200 | $4.50/GPU-hr | $2.45/GPU-hr | — | ||
| 7 | CoreWeave | NVIDIA HGX H200 (8x) 8× H200 · 128 vCPU · 2048 GB RAM | $6.31/GPU-hr | $2.62/GPU-hr | — | ||
| 8 | Oracle Cloud Infrastructure | H200 x1 1× H200 | $10.00/GPU-hr | — | — |
What is the H200 used for?
The H200 is a data-centre GPU positioned by its maker for large-scale model training, built on the Hopper architecture and released in 2023. 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 H200?
With 141 GB of memory, one card holds the weights of a model up to about 70 billion parameters at 16-bit precision, or about 141 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) | H200 (SXM, HGX) | H200 NVL |
|---|---|---|
| Architecture | Hopper | Hopper |
| Form factor | SXM (liquid-cooled) | PCIe FHFL dual-slot, passive |
| Memory | 141 GB HBM3e | 141 GB HBM3e |
| Memory bandwidth | 4,800 GB/s | 4,813 GB/s |
| Interconnect | NVLink 4th gen (900 GB/s) | NVLink bridge: 900 GB/s (2-way), 1,800 GB/s (4-way) |
| Max power (TDP) | 700 W | 600 W |
| Precisions listed | FP8, INT8, FP16, BF16, TF32 | FP8, INT8, FP16, BF16, TF32 |
| Partitioning | Up to 7 MIGs @ 18GB | Up to 7 MIGs @ 16.5GB |
| Released | 2023 | — |
| Source | maker spec sheet ↗ | maker spec sheet ↗ |
Need H200 capacity for a longer run?
Committed and cluster pricing is often negotiated. Tell us the GPU count, duration and region.