NVIDIA H200 rental price per hour
Series 1 · GPU prices · NVIDIA · snapshot 16 Oct 2026
The H200 rents for $4.40 per GPU-hour at the median — 26% more than the H100 for a chip NVIDIA rates at the same FP8 throughput but with 141 GB of memory instead of 80.
Figure 1H200: median rental price and published range, USD per GPU-hour, 2025-03-01 to 2026-10-14
§1Reading the H200 market
The cheapest recorded H200 hour is Beam's serverless rate of $2.09. That is a per-second serverless product, not a reserved machine, but it is a published on-demand price, so it sets the floor. The previous floor holder, FluidStack, no longer shows a public price page and appears to sell only by quote, so it dropped out of the record.
Above the floor the market bunches tightly. GMI Cloud lists $2.60, Jarvislabs $3.99 on demand (or $1.99 on spot), and the mid-tier group — Hyperbolic, Crusoe, DigitalOcean and Nebius — clusters between $4.0 and $4.6. GetDeploying's tracked median of $4.40 is the figure the index adopts. Azure is alone at the top: its ND96isr H200 v5 costs $110.24 an hour for eight GPUs, $13.78 each, and there is no reserved SKU for that instance.
Unlike the H100, the H200 median has edged up rather than down over the series: $3.79 in Mar 2025 against $4.40 now. The published 12-month rate — DigitalOcean at $3.40 — replaced an earlier unsourced estimate in the tracker and sits 23% under the on-demand median.
Table 1H200 prices by provider, USD per GPU-hour
| Provider | Offer | USD / GPU-hr | Detail | As of |
|---|---|---|---|---|
| Jarvislabs | Spot | $1.99 | 2026-09-21 | |
| Beam | Serverless | $2.09 | 2026-09-21 | |
| GMI Cloud | On-demand | $2.60 | 2026-09-21 | |
| DigitalOcean | Reserved, 12 mo | $3.40 | 2026-09-21 | |
| Jarvislabs | On-demand | $3.99 | 2026-09-21 | |
| GetDeploying median | Third-party index | $4.40 | 2026-09-21 | |
| Microsoft Azure | On-demand | $13.78 | ND96isr H200 v5, $110.24/hr for 8 GPUs | 2026-09-21 |
Rows are the prices recorded in the tracker note for this model; 4 of the 7 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, H200.
§2Specifications
Table 2H200 SXM (H200 NVL in brackets)
| Memory per GPU (index) | 141 GB |
| Memory | 141GB |
| Memory bandwidth | 4.8TB/s |
| Max TDP | Up to 700W, configurable (NVL: up to 600W) |
| FP8 Tensor Core, with sparsity | 3,958 TFLOPS (NVL: 3,341) |
| Interconnect | NVLink 900GB/s; PCIe Gen5 128GB/s |
| Form factor | SXM (NVL: PCIe dual-slot, air-cooled) |
Quoted from NVIDIA — H200 Tensor Core GPU, checked 17 Oct 2026. Vendor tensor figures are "with sparsity" unless stated; dense throughput is half.
§3What moves the H200 price
- Memory: NVIDIA lists 141GB at 4.8TB/s against the H100 SXM's 80GB at 3.35TB/s; a model too large for one 80 GB card may fit on one H200, which is what the premium buys.
- Identical compute rating: both SXM parts are listed at 3,958 FP8 teraFLOPS with sparsity, so for compute-bound training the H200 premium buys little.
- Thin hyperscaler competition: one Azure instance defines a ceiling 3.1× the median; the mid-tier clouds sit within about 60 cents of each other ($4.0–4.6).
- Reserved supply: only one provider in the record publishes a 12-month H200 rate, which keeps term pricing opaque.
§4Rent or buy
An eight-GPU H200 node at the median rate costs $25,696 per month of continuous use; with the DigitalOcean 12-month rate it is $19,856.
There is no cited H200 system price in our sources, so ownership economics cannot be computed honestly here. What can be said from the specs is that the H200 drops into the same SXM, 700 W-class board envelope as the H100, so a site already built for H100 nodes can usually host it — the decision is about the hardware quote, not new infrastructure.
Other models with their own page: GB200, B200, MI355X, H100, MI300X, A100, L40S, RTX 4090. All models: GPU rental price index. How the figures are built: methodology.