H200 · ways to pay
On-demand, spot and reserved H200 prices
6 providers list a spot or reserved H200 price in the data. The lowest on-demand price is $3.49 at Civo, the lowest spot is $2.58 at CoreWeave and the lowest reserved is $2.79 at Hyperstack. Different providers and sizes, so compare inside a row, not across them.
- On-demand low
- $3.49
- Spot low
- $2.58
- Reserved low
- $2.79
Civo
CoreWeave
Hyperstack
- Hyperstack
- DigitalOcean (Paperspace)
- Together AI
- CoreWeave
- Verda
- Google Cloud
By provider
One provider at a time
Each price shows its size. The percentage is against that provider's own lowest solid on-demand price, which is the only fair base.
| Provider | On-demand lowest | Spot lowest | Reserved lowest |
|---|---|---|---|
| Hyperstack | $3.991× SXM | none listed | $2.791× SXM, -30% |
| DigitalOcean (Paperspace) | $4.471× HGX | none listed | $3.401× HGX, -24% |
| Together AI | $5.998× HGX | $2.998× HGX, -50% | $4.158× HGX, -31% |
| CoreWeave | $6.308× HGX | $2.588× HGX, -59% | none listed |
| Verda | $6.501× SXM | $3.251× SXM, -50% | none listed |
| Google Cloud | $10.608× | $6.368×, -40% | none listed |
Not in this comparison: Nebius spot at $0.79 (8× HGX), listed as a from price. It is in the full list below.
Every row
All spot and reserved listings
Month = per GPU-hour × 720. Each row links to the page it was read from. A row marked "from price" is a lowest-offer figure the provider labels that way; it is shown here but kept out of the comparisons above.
| Provider | Type | Configuration | Per GPU-hour | Per 720-hour month | Read | Source |
|---|---|---|---|---|---|---|
| CoreWeave | spot | 8× HGX · NVIDIA HGX H200 | $2.58 | $1,857.60 | 2026-11-12 | page |
| CoreWeave | spot | 8× HGX · NVIDIA HGX H200 | $2.62 | $1,883.70 | 2026-11-12 | page |
| DigitalOcean (Paperspace) | reserved | 1× HGX · NVIDIA HGX H200 12-month reserved | $3.40 | $2,448.00 | 2026-11-12 | page |
| Google Cloud | spot | 8× · a3-ultragpu-8g (224 vCPU, 2,952GB RAM) | $6.36 | $4,578.40 | 2026-11-12 | page |
| Hyperstack | reserved | 1× SXM · NVIDIA H200 SXM reservation | $2.79 | $2,008.80 | 2026-11-12 | page |
| Nebius | spotfrom price, not ranked | 8× HGX · NVIDIA HGX H200 (preemptible) | $0.79 | $568.80 | 2026-11-12 | page |
| Together AI | reserved | 8× HGX · NVIDIA HGX H200 cluster reserved 31-90 days | $4.15 | $2,988.00 | 2026-11-12 | page |
| Together AI | reserved | 8× HGX · NVIDIA HGX H200 cluster reserved 3-30 days | $4.99 | $3,592.80 | 2026-11-12 | page |
| Together AI | spot | 8× HGX · NVIDIA HGX H200 cluster preemptible | $2.99 | $2,152.80 | 2026-11-12 | page |
| Verda | spot | 1× SXM · 1x H200 SXM5 141GB (44 vCPU, 170 GB RAM) spot | $3.25 | $2,341.44 | 2026-11-12 | page |
Ours, live
Runcrate prepaid terms, H200 NVL
From Runcrate's live GPU feed, Worcester, United Kingdom. We run Runcrate, so this is listed on the same terms as the rest. Runcrate pricing page
| Term | Per GPU-hour | Discount stated | On-demand reference | Per 720-hour month |
|---|---|---|---|---|
| 1 monthours, live | $4.44 | 5% | $4.67 | $3,194.28 |
| 3 monthsours, live | $4.30 | 8% | $4.67 | $3,093.41 |
| 6 monthsours, live | $4.16 | 11% | $4.67 | $2,992.54 |
| 12 monthsours, live | $3.97 | 15% | $4.67 | $2,858.04 |
Prepaid means paying for the term up front. The month column is the per-GPU-hour price × 720, our arithmetic, not a figure from the feed.
Definitions
Three ways to pay
- On-demand
- You start and stop when you like and pay for the time used. This is the only type counted in the ledger statistics.
- Spot or preemptible
- Spare capacity at a lower rate that the provider can take back. Nebius calls it preemptible; the data records it as spot.
- Reserved or prepaid
- You commit to a term and get a lower rate. Terms run from days to a year; where a page does not state the term, we say so.
Limits
What the data cannot say
- Availability. A listed price is not a promise that GPUs are free when you ask.
- Interruption rates. Nothing here says how often a spot instance is reclaimed; Nebius's own page says its preemptible prices can change as often as every 15 minutes.
- Term length, where the page leaves it out (Hyperstack). A reserved price without a stated term cannot be compared with one that has it.
- Volume discounts and contracts behind a sales call. Only published per-GPU-hour prices are held.
- Prices as of 12 November 2026. This is one snapshot, with no history.
Questions
Terms, answered
- How much lower is spot than on-demand?
- Within each provider's own prices: Together AI -50% ($2.99 against $5.99), CoreWeave -59% ($2.58 against $6.30), Verda -50% ($3.25 against $6.50), and Google Cloud -40% ($6.36 against $10.60). No provider lists spot above its own on-demand price.
- How much does a reservation save?
- Against each provider's lowest on-demand price: Hyperstack -30% ($2.79, NVIDIA H200 SXM reservation), DigitalOcean (Paperspace) -24% ($3.40, NVIDIA HGX H200 12-month reserved), and Together AI -31% ($4.15, NVIDIA HGX H200 cluster reserved 31-90 days). Term lengths differ and some are not stated, so the percentages are not like for like.
- Which provider has the lowest spot price?
- CoreWeave at $2.58 per GPU-hour (8× HGX, NVIDIA HGX H200), read 2026-11-12. The page says spot prices can change; a figure read on one day is not a standing offer. Nebius lists a lower from price of $0.79; it is shown in the table but kept out of the ranking.
- Does Runcrate offer prepaid terms for the H200?
- Its live feed lists H200 NVL terms of 1 month at $4.44 per GPU-hour, 3 months at $4.30 per GPU-hour, 6 months at $4.16 per GPU-hour, and 12 months at $3.97 per GPU-hour, against an on-demand reference of $4.67. They are in the Runcrate table above, marked "ours, live".
- What can this page not tell me?
- Whether capacity is free when you ask, how often a spot instance is interrupted, how long a reservation lasts where the page does not say, and what volume discounts exist behind a sales call. The data holds only published per-GPU-hour prices.
Disclosure
H200 Price Ledger is published by the Runcrate team, who also run Runcrate, an AI cloud. Runcrate is listed in the table for 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.