What is an NVIDIA H100?
The H100 data center GPU: variants, specifications and current cloud prices per GPU-hour.
Published Oct 9, 2026
Overview
The NVIDIA H100 is a data center GPU of the Hopper generation. It is used to train and run large AI models and is one of the most widely offered GPUs in the cloud.
In AI Compute Prices, 28 providers list H100 capacity in 404 offers, as of Oct 25, 2026.
Variants
H100 SXM is mounted on a server board, usually eight GPUs per server connected by NVLink. Clouds often list it as “8x HGX” or “H100 SXM”.
H100 NVL is a PCIe card with more memory. Clouds also list PCIe versions as “H100 PCIe”.
| Model | Memory | Memory bandwidth | FP8 Tensor | Max. power |
|---|---|---|---|---|
| H100 SXM | 80 GB | 3.35 TB/s | 3,958 TFLOPS | 700 W |
| H100 NVL | 94 GB | 3.9 TB/s | 3,341 TFLOPS | 350–400 W |
Manufacturer figures. Tensor Core values with sparsity, H200 values preliminary according to NVIDIA.
Cloud prices
On-demand prices for the H100 range from USD 2.64EUR 2.36CHF 2.18 to USD 13.27EUR 11.86CHF 10.94 per GPU-hour. The median across on-demand offers is USD 3.75EUR 3.35CHF 3.09.
Prices differ by variant, configuration (single GPU or eight-GPU server), pricing model and location.
As of Oct 25, 2026, 21:05 UTCSource: public price lists and pricing APIs of the providersConverted at ECB reference rates of Oct 23, 2026Method