NVIDIA H100 rental price per hour
Series 1 · GPU prices · NVIDIA · snapshot 16 Oct 2026
Renting one H100 80GB costs $3.49 an hour at the median published rate, but the offers on the market run from $1.30 to $12.29. Which end you land on depends far more on the kind of seller than on the chip.
Figure 1H100: median rental price and published range, USD per GPU-hour, 2023-06-01 to 2026-10-16
§1Reading the H100 market
Three tiers of seller explain almost the entire 9.5× spread. GPU marketplaces and pooled capacity sit at the bottom: Lium.io's pool at $1.30 is the floor, though the tracker flags that it was not re-checked in the latest cycle. Specialist GPU clouds form the middle — Runpod charges $2.69 on Community Cloud and $3.99 on Secure Cloud, Lambda lists $3.99–$4.29 for H100 SXM, Nebius $4.50. Hyperscalers set the ceiling: AWS ~$6.88, Google Cloud ~$11.00, and Azure's ND96isr H100 v5 at $12.29 once its $98.32 hourly rate is split across eight GPUs.
The direction changed this year. The median fell from $8.00 in Jun 2023 to a low of $1.95 in Oct 2025, the bottom of the long post-shortage slide, and today's $3.49 is +79% against the 1 Oct 2025 reading. The latest snapshot moved the median from $3.29 to $3.49 on 16 Oct 2026 because first-party price lists went up: Runpod's Secure Cloud H100 SXM rose from $3.29 to $3.99, and Nebius moved its HGX H100 from $3.85 to $4.50 effective 1 October 2026.
Term pricing barely helps at the moment. DigitalOcean's 12-month reserved rate of $3.26 is only 7% below the on-demand median, and it is above Runpod's Community Cloud rate. Buyers who can tolerate a marketplace host get a lower hourly figure without any commitment; buyers who need an SLA pay the middle tier either way.
Table 1H100 prices by provider, USD per GPU-hour
| Provider | Offer | USD / GPU-hr | Detail | As of |
|---|---|---|---|---|
| Lium.io | Marketplace | $1.30 | pool price; not re-verified in the 2026-10-16 cycle | 2026-09-21 |
| Runpod | On-demand | $2.69 | Community Cloud | 2026-10-16 |
| DigitalOcean | Reserved, 12 mo | $3.26 | 2026-09-28 | |
| Runpod | On-demand | $3.99 | Secure Cloud, H100 SXM (was $3.29) | 2026-10-16 |
| Lambda | On-demand | $3.99–$4.29 | H100 SXM | 2026-10-16 |
| Nebius | On-demand | $4.50 | HGX H100, raised from $3.85 effective 1 Oct 2026 | 2026-10-16 |
| AWS | On-demand | ~$6.88 | 2026-10-16 | |
| Google Cloud | On-demand | ~$11.00 | 2026-10-16 | |
| Microsoft Azure | On-demand | $12.29 | ND96isr H100 v5, $98.32/hr for 8 GPUs | 2026-10-16 |
Rows are the prices recorded in the tracker note for this model; 6 of the 9 come from the provider's own price page, the rest through comparison trackers. Third-party index rows are shown for context and not counted as providers. Every page behind this table: Sources, H100.
§2Specifications
Table 2H100 SXM (H100 NVL in brackets)
| Memory per GPU (index) | 80 GB |
| Memory | 80GB (NVL: 94GB) |
| Memory bandwidth | 3.35TB/s (NVL: 3.9TB/s) |
| Max TDP | Up to 700W, configurable (NVL: 350-400W) |
| FP8 Tensor Core, with sparsity | 3,958 teraFLOPS (NVL: 3,341) |
| Interconnect | NVLink 900GB/s; PCIe Gen5 128GB/s |
| Form factor | SXM (NVL: PCIe dual-slot, air-cooled) |
Quoted from NVIDIA — H100 Tensor Core GPU, checked 17 Oct 2026. Vendor tensor figures are "with sparsity" unless stated; dense throughput is half.
§3What moves the H100 price
- Seller type: marketplace pools, GPU clouds and hyperscalers price the same 80 GB part in three separate bands.
- Form factor: listings mix SXM boards (900GB/s NVLink, up to 700W per NVIDIA) and PCIe NVL cards (350–400W); check which one a price refers to before comparing.
- Supply of newer parts: the H200 median is $4.40 (+26% against the H100) for the same FP8 compute rating and more memory — that gap caps how far H100 rates can climb.
- Contract length: published 12-month discounts on H100 are thin (−7%), so negotiation on volume matters more than the term itself.
§4Rent or buy
At the median, one H100 rented around the clock costs about $2,548 a month, and a standard eight-GPU node $20,382. At the floor the same node is $7,592; at the Azure ceiling it is $71,774.
NVIDIA does not publish a price list for H100 servers and we have no dated purchase quote to cite, so this page prints no purchase price. If you are weighing ownership, the useful comparison is a supplier's quote for a complete eight-GPU system against the monthly node figures above, plus power and colocation — the form below returns real quotes.
Other models with their own page: GB200, B200, MI355X, H200, MI300X, A100, L40S, RTX 4090. All models: GPU rental price index. How the figures are built: methodology.