8x H200 cloud pricing
Every 8-GPU H200 node we track across 12 providers — from $15.97/hr ($2.00/GPU-hr).
Cheapest in-stock 8x H200 right now
LimitedWhat an 8x H200 node costs, and what is launchable
An 8x H200 node is billed as one server, so the $15.97/hr floor is $2.00 per GPU-hour. The median of the 40 priced 8x H200 listings is $43.20/hr ($5.40/GPU-hr) and the priciest is $94.87/hr — a 5.9× spread across 12 providers for the same 8 GPUs, which is why the provider matters more than the card.
4 of the 12 providers reported 8x H200 capacity on our last hourly check. The cheapest node confirmed in stock is $28.72/hr on Runpod (Global), 80% above the listed floor — the gap between the advertised price and what you can launch right now.
Spot or interruptible 8x H200 capacity starts at $6.32/hr (60% below the on-demand floor) from 7 providers — a fit for checkpointed training, not for serving. 5 of the 12 providers include egress (PrimeIntellect, Vast, Runpod, Hyperstack and 1 more); the rest charge from $0.01/GB, so pulling 1 TB of checkpoints out costs about $8.70 — roughly 1 hour of the cheapest 8x H200 node.
Before renting an 8x H200 node
Verify that the listed server matches the topology and networking your distributed workload needs.
- Displayed hourly rates are whole-node cloud rental rates for the 8-GPU server, not hardware purchase prices.
- Per-GPU cost is the whole-node price divided by eight; it is not a promise that one eighth of the node can be booked by itself.
- Confirm the GPU configuration, provider-stated intra-node topology and one-server placement, CPU/RAM, region, storage and egress charges before checkout.
- Treat interconnect and topology as unknown unless the provider source states them; do not infer NVLink, NVSwitch, InfiniBand, or multi-node networking from the H200 name or 8-GPU count.
- For multi-node jobs, verify the provider's cluster networking separately from the single-node listing.
8x H200 rental prices by provider
- datacrunch__8xH2…GB__176__1450Out100% · 30d$15.97/hr8× H200176 vCPU1450 GBFree egressFI, IS
- n1-standard-96 + 8xH200No data$79.05/hrSpot $47.29−40%8× H20096 vCPU360 GB$0.12/GB egressus-central1, us-west1
H200 (8× GPU) availability — last 7 days
1 of 8 providers kept 8× H200 ≥80% available all week. Polled hourly · 8 providers tracked · 7-day window.
| Provider | Sun4 | Mon5 | Tue6 | Wed7 | Thu8 | Fri9 | Sat10 | Week |
|---|---|---|---|---|---|---|---|---|
| PrimeIntellect | 99% | |||||||
| Shadeform | 15% | |||||||
| Nebius | 6% | |||||||
| Hyperstack | 0% | |||||||
| Lium | 0% | |||||||
| Runpod | 0% | |||||||
| Vast | 0% | |||||||
| Verda | 0% |
Dated research snapshot — separate from hourly refreshed live data
Eight H200s: turn a GPU-hour headline into a node budget
Sources were accessed 26 September 2026, 22:33–22:36 UTC. These are advertised rates, not a capacity reservation or checkout quote. All calculations below use USD.
An eight-GPU job needs an eight-GPU budget. In this two-provider sample, published on-demand GPU rates translate to $31.92 and $36.72 per eight-GPU hour. Those are budgeting calculations, not evidence that two identical servers are available. The more consequential distinction is the purchase model: an attractive “from” price can belong to spot capacity.
Two published offers, with the missing information kept visible
| Provider and product | GPU count | Published on-demand unit | Calculated eight-GPU compute budget | Region evidence | Availability at observation |
|---|---|---|---|---|---|
Hyperstack H200 SXM, flavor n3-H200-SXM5x8 | 8 | $3.99/GPU-hour | $31.92/hour; $766.08 for 24 hours | Flavor catalog: CANADA-1, Quebec, Canada | Not checked in authenticated stock/checkout; a catalog entry does not reserve capacity |
| Verda H200 GPU Instances; product page offers 1x, 2x, 4x and 8x | 8 selected for this calculation | $4.59/GPU-hour | $36.72/hour; $881.28 for 24 hours | This rate is not tied to a region in the product table. Location docs list FIN-01/02/03, Helsinki; exact 8x location requires confirmation | Not checked in authenticated stock/checkout; exact flavor and regional quote unverified |
Sources: Hyperstack pricing, Hyperstack flavor catalog, Verda H200 product table, Verda pricing, Verda locations.
Hyperstack's catalog identifies 176 CPUs, 1,800 GB system RAM, a 100 GB root disk and 48,000 GB ephemeral disk for this flavor. Temporary disk capacity is not a durable checkpoint plan. Verda's H200 product table explicitly mentions NVLink for GPU Instances, but that does not establish the exact topology or networking of a booked node. We have not established equivalent CPU, RAM, storage or interconnect configurations between these offers.
The price label can matter more than the small hourly difference
Verda's H200 headline starts at $2.30/hour, while its product table assigns $2.30 to spot and $4.59 to on-demand per GPU-hour. Multiplying the spot number by eight gives $18.40/hour, or $441.60 over 24 hours if that rate and capacity persist. Treating it as on-demand would underbudget the published eight-GPU on-demand calculation by $439.68 per day. This is a price-label comparison, not an estimate of spot job savings. Restart and lost-work costs are not measured. Source: Verda H200 pricing table.
The two on-demand calculations differ by $115.20 per 24 hours. That difference is only useful after location, usable capacity and workload fit are settled. A lower hourly price does not prove a lower cost per completed job.
Choose a launchable configuration, then compare the bill
For a Canada requirement, Hyperstack supplies a documented regional candidate; confirm eight-GPU stock and the final configuration. For a Finland requirement, investigate Verda's exact eight-GPU region and quote before using its headline rate in a budget. Verda says prices can vary by location. Neither observation proves immediate availability. Verda billing guidance.
Use the current 8x H200 provider table to shortlist providers and inspect each listing's region, stock status and observation time. Then follow that listing's provider action and verify the actual checkout quote. If eight GPUs are unnecessary, compare other H200 configurations. No provider wins this sample on price alone.
Method and limits
We read public provider pricing and configuration pages, selected on-demand GPU Instances where eight GPUs are explicitly offered, and multiplied the advertised GPU-hour rate by 8 and then 24. Dividing a node price by 8 is a comparison unit, not permission to rent one eighth of a node. Figures exclude unverified taxes and additional charges; confirm storage, IP, networking and billing conditions at checkout. Reserved, serverless and multi-node cluster products are excluded from the main table. The spot calculation is a separate illustration.
This is a deliberately small sample, not a market minimum, benchmark or inventory survey. No authenticated capacity API, rental, performance test or negotiated quote was used. Exact intra-node topology and multi-node connectivity remain unverified unless explicitly scoped above. Recheck the sources before booking; this dated comparison does not replace the live listings.
8x H200 pricing FAQ
How much does an 8x H200 server cost per hour?
Right now an 8-GPU H200 node starts at $15.97/hr ($2.00 per GPU-hour) across 12 providers, and the priciest 8x H200 listing we track is $94.87/hr. Prices are on-demand rates as published by each provider and refresh hourly.
How much does an 8x H200 cost per month?
At the current floor of $15.97/hr, running an 8x H200 node around the clock for 30 days costs about $11498.40 before egress and storage. Per-second and per-minute billing (see each provider page) avoids paying for partial hours on shorter jobs.
Which provider has 8x H200 nodes in stock right now?
On our last hourly check the cheapest 8x H200 node reported in stock was $28.72/hr on Runpod (Global). The table below marks every 8x H200 listing In stock, Limited, Out or No data, and the heatmap shows each provider's 8x availability over the last 7 days.