How much does an H100 cost per hour? 64 providers, one table.
Every published on-demand and spot price we could find for the NVIDIA H100 SXM, recorded 15 September 2026 with the source linked on each row, sorted from cheapest to most expensive. Prices are in US dollars per GPU-hour, storage and taxes excluded. Ours is in the table too, and it is the row we are judged on.
- Cheapest on-demand
- $1.15/hCompute Cheap, published rate
- Median on-demand
- $2.50/hAcross 64 providers
- Cheapest spot seen
- $0.69/hVast.ai, interruptible
- OnlineGPU
- $0.89/hOn-demand · spot $0.55 · reserved $0.65
Between $1.15 and $11.06 an hour, and the spread is not noise.
The same chip costs 10 times more at the top of the table than at the bottom. What you pay for beyond the GPU is the contract: enterprise support, bundled networking, capacity guarantees, or none of those.
- On-demand, cheapest
- $1.15 per GPU-hourCompute Cheap · $839.50 for a 730-hour month
- On-demand, median
- $2.50 per GPU-hourHalf of the 64 providers charge more, half charge less · $1,825.00 a month
- Spot or interruptible
- $0.69 per GPU-hour at Vast.aiCapacity the provider can take back; the notice period and what happens to the disk vary
- OnlineGPU
- $0.89 on-demand, $0.55 spot, $0.65 reservedBilled per minute; $401.50 a month on spot · 23% below the cheapest published on-demand rate
One month, one H100
730 h- Most expensive published rate$8,073.80
- Median of 64 providers$1,825.00
- Cheapest published rate$839.50
- OnlineGPU, on-demand$649.70
- OnlineGPU, spot$401.50
- Spot here against the median$1,423.50
Compute only. Storage is $0.08 per GB-month here and there are no egress fees; both vary elsewhere and can add a third to a bill.
H100 SXM, on-demand price per GPU-hour, 64 providers.
Lowest published rate we recorded for each provider, cheapest first. Spot is shown when the provider publishes one. The source is linked on every row; a provider with a comparison page here is linked to it.
| # | Provider | On-demandper GPU-hour | Spot | 730 h | Source |
|---|---|---|---|---|---|
| — | OnlineGPUthis site · how the price is set | $0.89 | $0.55 | $649.70 | pricing page |
| 1 | Compute Cheaprecorded 15 Sep 2026 | $1.15 | not published | $839.50 | computeprices.com |
| 2 | Liumrecorded 15 Sep 2026 | $1.30 | not published | $949.00 | getdeploying.com |
| 3 | Verdarecorded 15 Sep 2026 | $1.63 | $1.63 | $1,189.90 | shadeform.com |
| 4 | Latitude.shrecorded 15 Sep 2026 | $1.68 | not published | $1,226.40 | latitude.sh |
| 5 | Novita AIrecorded 15 Sep 2026 | $1.70 | $1.70 | $1,241.00 | getdeploying.com |
| 6 | Vast.airecorded 15 Sep 2026 | $1.73 | $0.69 | $1,262.90 | gpucloudcost.com |
| 7 | Gcorerecorded 15 Sep 2026 | $1.76 | not published | $1,284.80 | getdeploying.com |
| 8 | Cudo Computerecorded 15 Sep 2026 | $1.79 | not published | $1,306.70 | computeprices.com |
| 9 | HyperAIrecorded 15 Sep 2026 | $1.80 | not published | $1,314.00 | getdeploying.com |
| 10 | Beamrecorded 15 Sep 2026 | $1.83 | not published | $1,335.90 | getdeploying.com |
| 11 | UpCloudrecorded 15 Sep 2026 | $1.88 | $1.88 | $1,372.40 | getdeploying.com |
| 12 | Fal.airecorded 15 Sep 2026 | $1.89 | not published | $1,379.70 | comparegpuclouds.com |
| 13 | GPU Outletrecorded 15 Sep 2026 | $1.95 | not published | $1,423.50 | computeprices.com |
| 14 | GPU.airecorded 15 Sep 2026 | $1.98 | not published | $1,445.40 | getdeploying.com |
| 15 | Together AIrecorded 15 Sep 2026 | $1.99 | $1.99 | $1,452.70 | getdeploying.com |
| 16 | Voltage Parkrecorded 15 Sep 2026 | $1.99 | not published | $1,452.70 | voltagepark.com |
| 17 | GMI Cloudrecorded 15 Sep 2026 | $2.00 | not published | $1,460.00 | computeprices.com |
| 18 | Spheronrecorded 15 Sep 2026 | $2.05 | $2.05 | $1,496.50 | getdeploying.com |
| 19 | Nebiusrecorded 15 Sep 2026 | $2.15 | $2.15 | $1,569.50 | shadeform.com |
| 20 | Seewebrecorded 15 Sep 2026 | $2.18 | not published | $1,591.40 | computeprices.com |
| 21 | 1Legionrecorded 15 Sep 2026 | $2.20 | not published | $1,606.00 | computeprices.com |
| 22 | IO.NETrecorded 15 Sep 2026 | $2.20 | not published | $1,606.00 | io.net |
| 23 | TensorDockrecorded 15 Sep 2026 | $2.25 | $1.91 | $1,642.50 | tensordock.com |
| 24 | Vultrrecorded 15 Sep 2026 | $2.30 | $2.30 | $1,679.00 | getdeploying.com |
| 25 | Prime Intellectrecorded 15 Sep 2026 | $2.43 | $0.94 | $1,773.90 | primeintellect.ai |
| 26 | CoreWeaverecorded 15 Sep 2026 | $2.44 | $2.44 | $1,781.20 | coreweave.com |
| 27 | Denvr Dataworksrecorded 15 Sep 2026 | $2.45 | not published | $1,788.50 | cloud-gpus.com |
| 28 | Massed Computerecorded 15 Sep 2026 | $2.45 | $2.45 | $1,788.50 | vm.massedcompute.com |
| 29 | Civorecorded 15 Sep 2026 | $2.49 | not published | $1,817.70 | getdeploying.com |
| 30 | Lyceumrecorded 15 Sep 2026 | $2.49 | $1.10 | $1,817.70 | getdeploying.com |
| 31 | Hyperstackrecorded 15 Sep 2026 | $2.50 | $2.00 | $1,825.00 | hyperstack.cloud |
| 32 | Koyebrecorded 15 Sep 2026 | $2.50 | not published | $1,825.00 | getdeploying.com |
| 33 | Shadeformrecorded 15 Sep 2026 | $2.50 | not published | $1,825.00 | computeprices.com |
| 34 | Zettabyterecorded 15 Sep 2026 | $2.51 | not published | $1,832.30 | computeprices.com |
| 35 | Akash Networkrecorded 15 Sep 2026 | $2.55 | not published | $1,861.50 | console.akash.network |
| 36 | RunPodrecorded 15 Sep 2026 | $2.59 | $1.99 | $1,890.70 | runpod.io |
| 37 | Jarvis Labsrecorded 15 Sep 2026 | $2.69 | $1.19 | $1,963.70 | computeprices.com |
| 38 | Runcraterecorded 15 Sep 2026 | $2.75 | not published | $2,007.50 | getdeploying.com |
| 39 | Sestercerecorded 15 Sep 2026 | $2.75 | not published | $2,007.50 | getdeploying.com |
| 40 | EmpirioLabs AIrecorded 15 Sep 2026 | $2.90 | not published | $2,117.00 | getdeploying.com |
| 41 | Atlas Cloudrecorded 15 Sep 2026 | $2.95 | not published | $2,153.50 | computeprices.com |
| 42 | OVHcloudrecorded 15 Sep 2026 | $2.99 | not published | $2,182.70 | getdeploying.com |
| 43 | Geoddrecorded 15 Sep 2026 | $3.00 | not published | $2,190.00 | getdeploying.com |
| 44 | Hyperbolicrecorded 15 Sep 2026 | $3.19 | not published | $2,328.70 | computeprices.com |
| 45 | Thunder Computerecorded 15 Sep 2026 | $3.20 | not published | $2,336.00 | getdeploying.com |
| 46 | Canopy Waverecorded 15 Sep 2026 | $3.25 | not published | $2,372.50 | getdeploying.com |
| 47 | Lambdarecorded 15 Sep 2026 | $3.29 | not published | $2,401.70 | siliconanalysts.com |
| 48 | Scalewayrecorded 15 Sep 2026 | $3.31 | not published | $2,416.30 | getdeploying.com |
| 49 | Cerebriumrecorded 15 Sep 2026 | $3.40 | not published | $2,482.00 | getdeploying.com |
| 50 | Oblivusrecorded 15 Sep 2026 | $3.50 | not published | $2,555.00 | getdeploying.com |
| 51 | AceCloudrecorded 15 Sep 2026 | $3.63 | not published | $2,649.90 | computeprices.com |
| 52 | DigitalOceanrecorded 15 Sep 2026 | $3.89 | not published | $2,839.70 | shadeform.com |
| 53 | Crusoerecorded 15 Sep 2026 | $3.90 | not published | $2,847.00 | getdeploying.com |
| 54 | Modalrecorded 15 Sep 2026 | $3.95 | not published | $2,882.77 | gpucloudcost.com |
| 55 | Envergerecorded 15 Sep 2026 | $4.00 | not published | $2,920.00 | getdeploying.com |
| 56 | AtmosComputerecorded 15 Sep 2026 | $4.33 | not published | $3,160.90 | computeprices.com |
| 57 | Leafcloudrecorded 15 Sep 2026 | $4.77 | not published | $3,482.10 | cloud-gpus.com |
| 58 | Replicaterecorded 15 Sep 2026 | $5.49 | not published | $4,007.70 | gpucloudcost.com |
| 59 | Paperspacerecorded 15 Sep 2026 | $5.95 | not published | $4,343.50 | gpucloudcost.com |
| 60 | Basetenrecorded 15 Sep 2026 | $6.50 | not published | $4,745.00 | comparegpuclouds.com |
| 61 | AWSrecorded 15 Sep 2026 | $6.88 | $2.56 | $5,022.40 | computeprices.com |
| 62 | Microsoft Azurerecorded 15 Sep 2026 | $6.98 | $1.29 | $5,095.40 | gpucloudcost.com |
| 63 | Oracle Cloudrecorded 15 Sep 2026 | $10.00 | not published | $7,300.00 | cloud-gpus.com |
| 64 | Google Cloudrecorded 15 Sep 2026 | $11.06 | $3.69 | $8,073.80 | cloud.google.com |
Published prices move. If a figure is out of date, send us the link and the row is corrected. Nothing in this table is estimated: a provider without a published price is simply not in it.
SXM, PCIe or NVL: three cards share the name.
Most price lists say "H100" and mean the SXM part in an 8-GPU HGX node. The PCIe card is cheaper and slower; the NVL pairs two cards with more memory. The table below is our catalogue, with the lowest published price seen elsewhere for each.
| Variant | Memory | Bandwidth | GPU-to-GPU | Cheapest elsewhere | Providers | Here, on-demand | Here, spot |
|---|---|---|---|---|---|---|---|
| H100 SXMHopper · 2022 | 80 GB HBM3 | 3,350 GB/s | NVLink | $1.15 | 64 | $0.89 | $0.55 |
| H100 PCIeHopper · 2022 | 80 GB HBM2e | 2,000 GB/s | PCIe | $1.49 | 17 | $0.69 | $0.45 |
| H100 NVLHopper · 2023 | 94 GB HBM3 | 3,938 GB/s | NVLink | $1.11 | 12 | $0.79 | $0.49 |
Side by side with the numbers: H100 SXM vs H100 PCIe · H200 vs H100 · H100 vs H200 vs B200.
Five things that change an H100 bill more than the GPU does.
- Form factor
- SXM in an HGX node, PCIe card, or NVL pair. The table above shows the gap between them at the same provider is often larger than the gap between providers.
- Node size
- Many hourly rates are an 8-GPU node divided by eight; renting one card can cost more per GPU. Here the price per GPU is the same at 1×, 2×, 4× and 8×.
- Tier
- On-demand, spot or interruptible, reserved or committed. Spot is where the discounts live, and where the definitions differ most: notice period, what happens to the disk, whether the instance restarts.
- What is bundled
- Storage, egress, public IPs and support are sometimes included and sometimes billed on top. Egress alone can add a third to a monthly bill at a hyperscaler; it is free here.
- Commitment
- Monthly or yearly terms buy a lower rate in exchange for paying whether the card runs or not. Our reserved tier is $0.65 an hour for one to twelve months; the break-even against on-demand is on the reserved page.
Rent or buy?
assumption: $30,000 per cardStreet prices quoted for a new H100 card in 2026 run from about $25,000 to $40,000 (IntuitionLabs). Take $30,000 for one card, before the server it sits in, power, cooling and the people who keep it running. The hours of rental that spend the same amount:
- At the market median, $2.50/h12,000 h
- At the cheapest published rate, $1.15/h26,087 h
- Here on-demand, $0.89/h33,708 h
- Here on spot, $0.55/h54,545 h
54,545 hours is 6.2 years of running around the clock. A card that is busy that long, that consistently, is rare; for everything else, renting by the minute costs less and stops when the job does.
H100 pricing, in detail
How much does an H100 cost per hour?
Published on-demand prices for an H100 SXM run from $1.15 to $11.06 per GPU-hour across the 64 providers in this index (September 2026), with a median of $2.50. Spot and interruptible capacity is cheaper: $0.69 at Vast.ai and $0.55 here. On OnlineGPU an H100 SXM is $0.89 on-demand and $0.55 on spot, billed per minute.
What does an H100 cost per month?
At 730 hours a month: $839.50 at the cheapest published on-demand rate, $1,825.00 at the median, $649.70 here on-demand, $401.50 here on spot and $474.50 on our reserved tier, which holds the hardware for the term.
Why do H100 prices differ so much between providers?
The hardware is the same chip, but the offer around it is not: hyperscalers bundle networking, support tiers and enterprise contracts; marketplaces resell idle capacity with fewer guarantees; some prices are for an 8-GPU node divided by eight, others for a single card. Egress fees, storage and minimum commitments also move the real bill. This table compares one thing only: the published price of one GPU for one hour.
Is spot capacity really the same H100?
On OnlineGPU, yes: the same nodes, images and network as on-demand, sold from capacity we can reclaim with a 2-minute notice. The instance is stopped, never deleted, and the disk survives. Over the last 30 days <5% of spot H100 instances were reclaimed. Other providers define spot differently; check the notice period and what happens to the disk.
Should I rent or buy an H100?
Assume a card costs $30,000 to buy, before the server, power, cooling and the people to run it. At our on-demand rate it takes 33,708 GPU-hours of use, about 3.8 years of running day and night, to spend the same amount; at the market median of $2.50 it takes 12,000 hours, about 1.4 years. Renting wins unless the card would be busy around the clock for years.
More guides: the cheapest providers for eight GPUs · H100 vs H200 vs B200 · every provider against us, model by model
An H100 SXM for $0.55 an hour, billed by the minute.
Prepaid in crypto, no card, no contract. Stop it whenever you want and the charge stops with it.
Billed per minute from the moment the instance is reachable. Minimum credit $30, no subscription.