H100 · H200
Read 12 November 2026
H100 against H200: price and memory
- H200 lowest, against H100
- +75%
- $3.49 against $1.99 per GPU-hour
- H100 median
- $3.90/GPU-hr
- 19 providers
- H200 median
- $5.40/GPU-hr
- 15 providers
- List both
- 15
- providers
Basis: each GPU's lowest price per provider on current on-demand rows the provider sets. The gap is (H200 price minus H100 price) divided by the H100 price. Marketplace figures and announced prices are left out. USD per GPU-hour, before tax.
Side by side
derived columns are labelled| Measure | H100 | H200 | H200 against H100 |
|---|---|---|---|
| Lowest $/GPU-hr | $1.99 | $3.49 | +75% |
| Median $/GPU-hr | $3.90 | $5.40 | +38% |
| Highest $/GPU-hr | $11.06 | $10.60 | -4% |
| Providers with a price | 19 | 15 | 15 list both |
| VRAM, GB (catalogue) | 80 | 141 | +76% |
| Lowest $ per GB-hour (derived) | $0.0249 | $0.0248 | -0% |
| Median $ per GB-hour (derived) | $0.0488 | $0.0383 | -21% |
Price per GB-hour = price per GPU-hour divided by the catalogue VRAM in GB: $1.99 / 80 = $0.0249 for the H100 and $3.49 / 141 = $0.0248 for the H200. Memory is only one of the things a GPU rental buys; this column says nothing about speed.
Memory as the maker states it
catalogue VRAM is used for the price per GB| GPU | Variant | Memory GB | Memory type | Bandwidth TB/s | Source |
|---|---|---|---|---|---|
| H100 | SXM | 80 | not stated | 3.35 | maker page |
| H100 | NVL | 94 | HBM3 | 3.9 | maker page |
| H200 | SXM | 141 | HBM3e | 4.8 | maker page |
| H200 | NVL | 141 | HBM3e | 4.8 | maker page |
Figures are copied from each maker page with the variant it names; where a page does not state a figure the cell says not stated. The price per GB above uses the catalogue VRAM (80 GB and 141 GB) so that every GPU in the data has one comparable number.
Providers that list both
15 providersAt 1 of 15 of these providers the H200 is the lower price. The gap from the H100 runs from -4% to +107%, with a median of +20% across the providers.
Runcrate's live rows used above
ours, liveH100
H100 or H200: common questions
Is the H200 lower priced than the H100?
Compare the lowest listings first: the H200 is $3.49 against $1.99 for the H100, 75% above the H100. On the median provider low it is $5.40 against $3.90, 38% above the H100.
Which costs less per GB of memory?
At the lowest listing the H200 is $0.0248 per GB-hour and the H100 is $0.0249, so the H200 is level with the H100 on that measure. At the median the figures are $0.0383 and $0.0488. This uses the catalogue VRAM of 141 GB and 80 GB.
How much memory does the H200 have compared with the H100?
141 GB against 80 GB in the catalogue, which is 76% above the H100 (1.76 times). The maker table above lists the stated figure for each variant.
Which providers list both?
15 providers: Hyperstack, Massed Compute, RunPod, Civo, OVHcloud, Runcrate, ours, Crusoe, and Together AI and 7 more.
Does the price gap hold at the same provider?
At 1 of 15 providers the H200 is lower than that provider's own H100 price. The gaps run from -4% to +107%, so the market-wide gap of +38% between medians is not the gap you will see at every provider.
Is Runcrate included in these numbers?
Yes, where it has a live price: H100 $3.58 and H200 $4.67 per GPU-hour. It is sorted by price like every other row and carries no pinned position.
Disclosure
H100 Rate Card is published by the Runcrate team, who also run Runcrate, an AI cloud. Runcrate is listed in the table for the H100 and H200, marked "ours, live". It is sorted by price like every other row, with no pinned position. Other providers' prices are read from their own public pricing pages. We have no paid placement.