Spot, preemptible and interruptible GPU prices
The cheapest spot price per GPU-hour from each of 8 cloud providers, for every GPU at least 2 of them price on spot. Each row uses the provider’s own term, shows how the price is set, its discount against the same listing’s on-demand price, and what happens when the capacity is reclaimed. Ordered by price, cheapest first.
Disclosure: GPU Finder's founder, Alexander Salikov, works at Nebius. Nebius is ranked by observed price like every other provider, with no referral link or paid placement.
A100-80GB spot prices
6 providers · cheapest $0.13/GPU-hr on Vast. A100-80GB on-demand prices
| Provider | Term · how set | Spot / GPU-hr | On-demand / GPU-hr | Discount | If interrupted |
|---|---|---|---|---|---|
| Vast2× · Hong Kong, HK | Interruptible | $0.13updated daily · as of | $0.67 | 81% | Paused; resumes when it regains priority |
| Azure1× · westus3 | Spot | $0.40updated daily · as of | $3.67 | 89% | Deallocated by default; restart it yourself |
| Verda1× · FIN-02 | Spot | $1.01updated daily · as of | $2.02 | 50% | Not documented |
| Hyperstack1× · CANADA-1 | Spot | $1.08updated daily · as of | $1.35 | 20% | Deleted without notice; redeploy |
| AWS8× · ap-northeast-1 | Spot | $1.73updated daily · as of | $4.70 | 63% | Terminated by default (stop or hibernate on request) |
| Google Cloud8× · us-east4 | Spot | $2.02updated daily · as of | $5.07 | 60% | Stopped by default; restart it yourself |
A100 spot prices
5 providers · cheapest $0.22/GPU-hr on Vast. A100 on-demand prices
| Provider | Term · how set | Spot / GPU-hr | On-demand / GPU-hr | Discount | If interrupted |
|---|---|---|---|---|---|
| Vast1× · South Korea, KR | Interruptible | $0.22updated daily · as of | $0.74 | 69% | Paused; resumes when it regains priority |
| Azure8× · westus2 | Spot | $0.37updated daily · as of | $3.40 | 89% | Deallocated by default; restart it yourself |
| AWS8× · us-east-2 | Spot | $0.57updated daily · as of | $2.74 | 79% | Terminated by default (stop or hibernate on request) |
| Verda1× · FIN-02 | Spot | $0.73updated daily · as of | $1.46 | 50% | Not documented |
| Google Cloud8× · us-west4 | Spot | $1.96updated daily · as of | $3.74 | 48% | Stopped by default; restart it yourself |
H100 spot prices
5 providers · cheapest $0.67/GPU-hr on Vast. H100 on-demand prices
| Provider | Term · how set | Spot / GPU-hr | On-demand / GPU-hr | Discount | If interrupted |
|---|---|---|---|---|---|
| Vast1× · South Africa, ZA | Interruptible | $0.67updated daily · as of | $2.25 | 70% | Paused; resumes when it regains priority |
| Nebius1× · eu-north1 | Preemptible | $0.79$0.79$4.49updated hourly · as of | $4.50 | 82% | Stopped; restart it yourself (Managed Kubernetes restarts nodes) |
| Google Cloud8× · europe-west2 | Spot | $1.91updated daily · as of | $11.90 | 84% | Stopped by default; restart it yourself |
| Hyperstack1× · CANADA-1 | Spot | $2.00updated daily · as of | $2.50 | 20% | Deleted without notice; redeploy |
| AWS8× · ca-central-1 | Spot | $2.20updated daily · as of | $7.91 | 72% | Terminated by default (stop or hibernate on request) |
H200 spot prices
5 providers · cheapest $0.79/GPU-hr on Nebius. H200 on-demand prices
| Provider | Term · how set | Spot / GPU-hr | On-demand / GPU-hr | Discount | If interrupted |
|---|---|---|---|---|---|
| Nebius1× · eu-west1 | Preemptible | $0.79updated hourly · as of | $5.40 | 85% | Stopped; restart it yourself (Managed Kubernetes restarts nodes) |
| Vast1× · Quebec, CA | Interruptible | $1.33updated daily · as of | $3.60 | 63% | Paused; resumes when it regains priority |
| AWS8× · eu-north-1 | Spot | $2.16updated daily · as of | $8.47 | 75% | Terminated by default (stop or hibernate on request) |
| Verda4× · FIN-03 | Spot | $2.75updated daily · as of | $5.49 | 50% | Not documented |
| Google Cloud8× · us-east4 | Spot | $5.29updated daily · as of | $9.95 | 47% | Stopped by default; restart it yourself |
RTX PRO 6000 spot prices
5 providers · cheapest $0.47/GPU-hr on Vast. RTX PRO 6000 on-demand prices
| Provider | Term · how set | Spot / GPU-hr | On-demand / GPU-hr | Discount | If interrupted |
|---|---|---|---|---|---|
| Vast2× · Texas, US | Interruptible | $0.47updated daily · as of | $1.68 | 72% | Paused; resumes when it regains priority |
| Google Cloud8× · us-east1 | Spot | $0.58updated daily · as of | $1.67 | 65% | Stopped by default; restart it yourself |
| AWS1× · eu-north-1 | Spot | $0.76updated daily · as of | $7.60 | 90% | Terminated by default (stop or hibernate on request) |
| Verda4× · FIN-01 | Spot | $1.16updated daily · as of | $2.33 | 50% | Not documented |
| Hyperstack1× · CANADA-1 | Spot | $1.48updated daily · as of | $1.85 | 20% | Deleted without notice; redeploy |
V100 spot prices
5 providers · cheapest $0.08/GPU-hr on Vast. V100 on-demand prices
| Provider | Term · how set | Spot / GPU-hr | On-demand / GPU-hr | Discount | If interrupted |
|---|---|---|---|---|---|
| Vast1× · Hungary, HU | Interruptible | $0.08updated daily · as of | $0.11 | 27% | Paused; resumes when it regains priority |
| Verda2× · FIN-01 | Spot | $0.11updated daily · as of | $0.22 | 50% | Not documented |
| Azure8× · eastus | Spot | $0.30updated daily · as of | $2.75 | 89% | Deallocated by default; restart it yourself |
| AWS8× · us-east-1 | Spot | $0.43updated daily · as of | $3.90 | 89% | Terminated by default (stop or hibernate on request) |
| Google Cloud8× · asia-east1 | Spot | $0.45updated daily · as of | $3.14 | 86% | Stopped by default; restart it yourself |
B200 spot prices
4 providers · cheapest $0.99/GPU-hr on Nebius. B200 on-demand prices
| Provider | Term · how set | Spot / GPU-hr | On-demand / GPU-hr | Discount | If interrupted |
|---|---|---|---|---|---|
| Nebius1× · us-central1 | Preemptible | $0.99$0.99$8.49updated hourly · as of | $8.50 | 88% | Stopped; restart it yourself (Managed Kubernetes restarts nodes) |
| AWS8× · eu-north-1 | Spot | $1.88updated daily · as of | $17.91 | 89% | Terminated by default (stop or hibernate on request) |
| Google Cloud8× · asia-south2 | Spot | $1.92updated daily · as of | $11.96 | 84% | Stopped by default; restart it yourself |
| Vast8× · California, US | Interruptible | $6.25updated daily · as of | $9.38 | 33% | Paused; resumes when it regains priority |
L4 spot prices
3 providers · cheapest $0.13/GPU-hr on Vast. L4 on-demand prices
| Provider | Term · how set | Spot / GPU-hr | On-demand / GPU-hr | Discount | If interrupted |
|---|---|---|---|---|---|
| Vast4× · Pennsylvania, US | Interruptible | $0.13updated daily · as of | $0.20 | 33% | Paused; resumes when it regains priority |
| AWS1× · eu-north-1 | Spot | $0.15updated daily · as of | $1.04 | 86% | Terminated by default (stop or hibernate on request) |
| Google Cloud8× · asia-south1 | Spot | $0.53updated daily · as of | $1.27 | 58% | Stopped by default; restart it yourself |
L40S spot prices
3 providers · cheapest $0.33/GPU-hr on Vast. L40S on-demand prices
| Provider | Term · how set | Spot / GPU-hr | On-demand / GPU-hr | Discount | If interrupted |
|---|---|---|---|---|---|
| Vast4× · Japan, JP | Interruptible | $0.33updated daily · as of | $0.47 | 29% | Paused; resumes when it regains priority |
| AWS4× · us-east-1 | Spot | $0.38updated daily · as of | $3.77 | 90% | Terminated by default (stop or hibernate on request) |
| Nebius1× · eu-north1 | Preemptible | $0.75updated hourly · as of | $1.55 | 52% | Stopped; restart it yourself (Managed Kubernetes restarts nodes) |
RTX A6000 spot prices
3 providers · cheapest $0.14/GPU-hr on Vast. RTX A6000 on-demand prices
| Provider | Term · how set | Spot / GPU-hr | On-demand / GPU-hr | Discount | If interrupted |
|---|---|---|---|---|---|
| Vast1× · Italy, IT | Interruptible | $0.14updated daily · as of | $0.36 | 61% | Paused; resumes when it regains priority |
| Verda1× · FIN-02 | Spot | $0.36updated daily · as of | $0.73 | 50% | Not documented |
| Hyperstack1× · CANADA-1 | Spot | $0.40updated daily · as of | $0.50 | 20% | Deleted without notice; redeploy |
T4 spot prices
3 providers · cheapest $0.06/GPU-hr on Azure. T4 on-demand prices
| Provider | Term · how set | Spot / GPU-hr | On-demand / GPU-hr | Discount | If interrupted |
|---|---|---|---|---|---|
| Azure1× · northcentralus | Spot | $0.06updated daily · as of | $0.53 | 89% | Deallocated by default; restart it yourself |
| AWS1× · eu-north-1 | Spot | $0.09updated daily · as of | $0.56 | 85% | Terminated by default (stop or hibernate on request) |
| Google Cloud1× · asia-southeast2 | Spot | $0.18updated daily · as of | $0.91 | 81% | Stopped by default; restart it yourself |
A10 spot prices
2 providers · cheapest $0.24/GPU-hr on Vast. A10 on-demand prices
| Provider | Term · how set | Spot / GPU-hr | On-demand / GPU-hr | Discount | If interrupted |
|---|---|---|---|---|---|
| Vast1× · Illinois, US | Interruptible | $0.24updated daily · as of | $0.24 | 1% | Paused; resumes when it regains priority |
| Azure1× · westus2 | Spot | $0.80updated daily · as of | $3.20 | 75% | Deallocated by default; restart it yourself |
B300 spot prices
2 providers · cheapest $3.60/GPU-hr on Nebius. B300 on-demand prices
| Provider | Term · how set | Spot / GPU-hr | On-demand / GPU-hr | Discount | If interrupted |
|---|---|---|---|---|---|
| Nebius1× · uk-south1 | Preemptible | $3.60updated hourly · as of | $9.50 | 62% | Stopped; restart it yourself (Managed Kubernetes restarts nodes) |
| AWS8× · eu-west-2 | Spot | $4.57updated daily · as of | $27.19 | 83% | Terminated by default (stop or hibernate on request) |
L40 spot prices
2 providers · cheapest $0.35/GPU-hr on Vast. L40 on-demand prices
| Provider | Term · how set | Spot / GPU-hr | On-demand / GPU-hr | Discount | If interrupted |
|---|---|---|---|---|---|
| Vast1× · Vietnam, VN | Interruptible | $0.35updated daily · as of | $0.36 | 5% | Paused; resumes when it regains priority |
| Hyperstack1× · CANADA-1 | Spot | $0.80updated daily · as of | $1.00 | 20% | Deleted without notice; redeploy |
RTX 3080 spot prices
2 providers · cheapest $0.06/GPU-hr on Salad. RTX 3080 on-demand prices
| Provider | Term · how set | Spot / GPU-hr | On-demand / GPU-hr | Discount | If interrupted |
|---|---|---|---|---|---|
| Salad1× · Global (automatic placement) | Interruptible | $0.06updated daily · as of | $0.14 | 57% | Not documented |
| Vast1× · Chongqing, CN | Interruptible | $0.08updated daily · as of | $0.08 | 3% | Paused; resumes when it regains priority |
RTX 3080 Ti spot prices
2 providers · cheapest $0.08/GPU-hr on Salad. RTX 3080 Ti on-demand prices
| Provider | Term · how set | Spot / GPU-hr | On-demand / GPU-hr | Discount | If interrupted |
|---|---|---|---|---|---|
| Salad1× · Global (automatic placement) | Interruptible | $0.08updated daily · as of | $0.15 | 48% | Not documented |
| Vast1× · Germany, DE | Interruptible | $0.12updated daily · as of | $0.13 | 11% | Paused; resumes when it regains priority |
RTX 3090 spot prices
2 providers · cheapest $0.07/GPU-hr on Vast. RTX 3090 on-demand prices
| Provider | Term · how set | Spot / GPU-hr | On-demand / GPU-hr | Discount | If interrupted |
|---|---|---|---|---|---|
| Vast1× · India, IN | Interruptible | $0.07updated daily · as of | $0.10 | 31% | Paused; resumes when it regains priority |
| Salad1× · Global (automatic placement) | Interruptible | $0.09updated daily · as of | $0.17 | 47% | Not documented |
RTX 3090 Ti spot prices
2 providers · cheapest $0.10/GPU-hr on Salad. RTX 3090 Ti on-demand prices
| Provider | Term · how set | Spot / GPU-hr | On-demand / GPU-hr | Discount | If interrupted |
|---|---|---|---|---|---|
| Salad1× · Global (automatic placement) | Interruptible | $0.10updated daily · as of | $0.19 | 47% | Not documented |
| Vast1× · Washington, US | Interruptible | $0.11updated daily · as of | $0.24 | 56% | Paused; resumes when it regains priority |
RTX 4060 Ti spot prices
2 providers · cheapest $0.08/GPU-hr on Salad. RTX 4060 Ti on-demand prices
| Provider | Term · how set | Spot / GPU-hr | On-demand / GPU-hr | Discount | If interrupted |
|---|---|---|---|---|---|
| Salad1× · Global (automatic placement) | Interruptible | $0.08updated daily · as of | $0.14 | 43% | Not documented |
| Vast2× · United States, US | Interruptible | $0.08updated daily · as of | $0.11 | 28% | Paused; resumes when it regains priority |
RTX 4070 spot prices
2 providers · cheapest $0.07/GPU-hr on Salad. RTX 4070 on-demand prices
| Provider | Term · how set | Spot / GPU-hr | On-demand / GPU-hr | Discount | If interrupted |
|---|---|---|---|---|---|
| Salad1× · Global (automatic placement) | Interruptible | $0.07updated daily · as of | $0.15 | 53% | Not documented |
| Vast4× · United States, US | Interruptible | $0.09updated daily · as of | $0.16 | 43% | Paused; resumes when it regains priority |
RTX 4070 Ti spot prices
2 providers · cheapest $0.07/GPU-hr on Vast. RTX 4070 Ti on-demand prices
| Provider | Term · how set | Spot / GPU-hr | On-demand / GPU-hr | Discount | If interrupted |
|---|---|---|---|---|---|
| Vast1× · , CN | Interruptible | $0.07updated daily · as of | $0.07 | 3% | Paused; resumes when it regains priority |
| Salad1× · Global (automatic placement) | Interruptible | $0.08updated daily · as of | $0.20 | 60% | Not documented |
RTX 4080 spot prices
2 providers · cheapest $0.11/GPU-hr on Salad. RTX 4080 on-demand prices
| Provider | Term · how set | Spot / GPU-hr | On-demand / GPU-hr | Discount | If interrupted |
|---|---|---|---|---|---|
| Salad1× · Global (automatic placement) | Interruptible | $0.11updated daily · as of | $0.23 | 52% | Not documented |
| Vast1× · California, US | Interruptible | $0.19updated daily · as of | $0.27 | 30% | Paused; resumes when it regains priority |
RTX 4090 spot prices
2 providers · cheapest $0.12/GPU-hr on Vast. RTX 4090 on-demand prices
| Provider | Term · how set | Spot / GPU-hr | On-demand / GPU-hr | Discount | If interrupted |
|---|---|---|---|---|---|
| Vast1× · Texas, US | Interruptible | $0.12updated daily · as of | $0.14 | 14% | Paused; resumes when it regains priority |
| Salad1× · Global (automatic placement) | Interruptible | $0.16updated daily · as of | $0.33 | 52% | Not documented |
RTX 5060 spot prices
2 providers · cheapest $0.07/GPU-hr on Salad. RTX 5060 on-demand prices
| Provider | Term · how set | Spot / GPU-hr | On-demand / GPU-hr | Discount | If interrupted |
|---|---|---|---|---|---|
| Salad1× · Global (automatic placement) | Interruptible | $0.07updated daily · as of | $0.11 | 41% | Not documented |
| Vast1× · Quebec, CA | Interruptible | $0.08updated daily · as of | $0.14 | 41% | Paused; resumes when it regains priority |
RTX 5060 Ti spot prices
2 providers · cheapest $0.04/GPU-hr on Vast. RTX 5060 Ti on-demand prices
| Provider | Term · how set | Spot / GPU-hr | On-demand / GPU-hr | Discount | If interrupted |
|---|---|---|---|---|---|
| Vast1× · Texas, US | Interruptible | $0.04updated daily · as of | $0.20 | 80% | Paused; resumes when it regains priority |
| Salad1× · Global (automatic placement) | Interruptible | $0.07updated daily · as of | $0.15 | 53% | Not documented |
RTX 5070 Ti spot prices
2 providers · cheapest $0.10/GPU-hr on Salad. RTX 5070 Ti on-demand prices
| Provider | Term · how set | Spot / GPU-hr | On-demand / GPU-hr | Discount | If interrupted |
|---|---|---|---|---|---|
| Salad1× · Global (automatic placement) | Interruptible | $0.10updated daily · as of | $0.23 | 56% | Not documented |
| Vast1× · South Korea, KR | Interruptible | $0.11updated daily · as of | $0.12 | 13% | Paused; resumes when it regains priority |
RTX 5080 spot prices
2 providers · cheapest $0.15/GPU-hr on Salad. RTX 5080 on-demand prices
| Provider | Term · how set | Spot / GPU-hr | On-demand / GPU-hr | Discount | If interrupted |
|---|---|---|---|---|---|
| Salad1× · Global (automatic placement) | Interruptible | $0.15updated daily · as of | $0.26 | 42% | Not documented |
| Vast1× · , CN | Interruptible | $0.20updated daily · as of | $0.20 | 1% | Paused; resumes when it regains priority |
RTX 5090 spot prices
2 providers · cheapest $0.25/GPU-hr on Salad. RTX 5090 on-demand prices
| Provider | Term · how set | Spot / GPU-hr | On-demand / GPU-hr | Discount | If interrupted |
|---|---|---|---|---|---|
| Salad1× · Global (automatic placement) | Interruptible | $0.25updated daily · as of | $0.50 | 50% | Not documented |
| Vast1× · Denmark, DK | Interruptible | $0.29updated daily · as of | $0.67 | 57% | Paused; resumes when it regains priority |
RTX 6000 Ada spot prices
2 providers · cheapest $0.27/GPU-hr on Vast. RTX 6000 Ada on-demand prices
| Provider | Term · how set | Spot / GPU-hr | On-demand / GPU-hr | Discount | If interrupted |
|---|---|---|---|---|---|
| Vast2× · Australia, AU | Interruptible | $0.27updated daily · as of | $0.67 | 60% | Paused; resumes when it regains priority |
| Verda1× · FIN-02 | Spot | $0.66updated daily · as of | $1.32 | 50% | Not documented |
RTX PRO 4500 spot prices
2 providers · cheapest $0.27/GPU-hr on Vast. RTX PRO 4500 on-demand prices
| Provider | Term · how set | Spot / GPU-hr | On-demand / GPU-hr | Discount | If interrupted |
|---|---|---|---|---|---|
| Vast1× · Japan, JP | Interruptible | $0.27updated daily · as of | $0.34 | 21% | Paused; resumes when it regains priority |
| AWS1× · eu-south-2 | Spot | $0.30updated daily · as of | $2.65 | 88% | Terminated by default (stop or hibernate on request) |
How spot GPU pricing works
Spot capacity is the provider’s idle GPUs sold at a discount on the condition that it can take them back. The grey badge next to each term says how the price is set:
- Fixed: a published flat rate below on-demand.
- Dynamic: the provider reprices it from supply and demand.
- Bid: the lowest price you can bid; higher bids can outbid and interrupt you.
- Marketplace: hosts set their own interruptible prices.
When the provider reclaims capacity the instance is stopped, deleted or paused depending on the provider (“If interrupted”); anything on local disk may be lost. Save checkpoints to storage that outlives the instance, and keep an on-demand fallback for anything with a deadline.
Methodology
Prices come from each provider’s API or published catalogue, re-read hourly for providers whose prices move within the day and daily for the rest (each row says which, and when it was last read). For every GPU and provider the table shows the cheapest spot listing per GPU-hour across regions and node sizes. A spot price counts only when it is at least 1% below the same listing’s on-demand price and no lower than 10% of it; lower values are bid-floor artefacts. Providers without a first-party spot price are not included, and GPUs priced on spot by fewer than 2 providers are not listed. Interruption rates are not comparable across providers, so nothing is ranked by them. Rows are ordered by price alone; no provider pays for placement.
Download the data: CSV · JSON. Weekly spot history per GPU is on each GPU’s price history page, e.g. H100 price history.
Frequently asked questions
What is the difference between spot, preemptible and interruptible GPUs?
They are different names for the same idea: spare GPU capacity sold below the on-demand price that the provider can take back when it needs the capacity. The term depends on the provider: spot (AWS, Azure, Google Cloud, Runpod, Verda, Hyperstack); preemptible (Nebius); interruptible (Vast, Salad). A Vast bid price is the lowest price you can bid; a higher bid can interrupt you.
Which provider has the cheapest H100 spot price?
Right now Vast at $0.67/GPU-hr (interruptible, 70% below its on-demand price), then Nebius at $0.79/GPU-hr. 5 providers price the H100 on spot; the table is re-read hourly and ordered by price only.
What is the difference between a maximum price and following the spot price?
Where spot prices move with demand you can usually cap what you pay. With a maximum price the instance is stopped or evicted once the spot price rises above it (AWS, Azure, Nebius). Without a cap — AWS's default cap is the on-demand price, Azure's max price −1, Nebius's "follow spot price" mode — you are never stopped for price, but the provider can still reclaim the capacity. GPU Finder shows the price you pay in the uncapped mode.
Which workloads suit spot GPUs?
Jobs that save progress and can restart: checkpointed training and fine-tuning, batch inference, evaluation sweeps, rendering and experiments. Avoid spot for production serving without an on-demand fallback, deadline-bound runs and jobs that cannot checkpoint.
Does the ranking favour any provider?
No. Rows are ordered by observed spot price per GPU-hour, every provider gets the same template and its own term, and there are no featured or paid slots. GPU Finder's founder works at Nebius; Nebius is ranked like every other provider, with no referral link or paid placement.