October 2026 edition · Price index
GPU rental price index
What it costs to rent one data-centre GPU by the hour, on demand: the cheapest price, the median and the highest across 5 GPU clouds (DigitalOcean GPU Droplets, Hyperstack, Lambda, RunPod and Vultr Cloud GPU). US dollars per GPU-hour, each price from the provider's own pricing page.
5 GPUs · 16 prices · checked Oct 17–18, 2026
The index
On-demand price per GPU-hour. One price per provider: its smallest on-demand size.
| GPU | Cheapest | Cheapest at | Cheapest single GPU | Median | Highest | Providers | Checked |
|---|---|---|---|---|---|---|---|
| B200 | $6.00 |  Hyperstack8 GPUs only | $6.99Lambda | $6.99 | $7.99 | 3 | Oct 17–18, 2026 |
| H200 | $3.99 | Hyperstack8 GPUs only | $4.47DigitalOcean GPU Droplets | $4.47 | $5.29 | 3 | Oct 17–18, 2026 |
| H100 SXM | $3.20 | Hyperstack8 GPUs only | $3.99RunPod | $4.14 | $4.41 | 4 | Oct 17–18, 2026 |
| A100 80GB PCIe | $1.35 | Hyperstack | $1.35Hyperstack | $1.79 | $2.40 | 3 | Oct 17–18, 2026 |
| L40S | $1.09 |  RunPod | $1.09RunPod | $1.57 | $1.67 | 3 | Oct 17–18, 2026 |
- B200$6.00 to $7.99 · median $6.99
- H200$3.99 to $5.29 · median $4.47
- H100 SXM$3.20 to $4.41 · median $4.14
- A100 80GB PCIe$1.35 to $2.40 · median $1.79
- L40S$1.09 to $1.67 · median $1.57
Not in the index yet
A GPU gets a row once we have prices from at least 3 providers. These have fewer so far, so we don't give a median for them:
- B300: Hyperstack and RunPod
- GH200: Lambda
- H100 PCIe: Hyperstack and Lambda
- A100 80GB SXM: Lambda
- AMD MI300X: DigitalOcean GPU Droplets
- A40: Vultr Cloud GPU
- RTX 6000 Ada: DigitalOcean GPU Droplets
- RTX 5090: RunPod
- RTX 4090: RunPod
- RTX 4000 Ada: DigitalOcean GPU Droplets
Consumer cards like the RTX 4090 and 5090 are mostly rented on marketplaces, which the index leaves out. Their prices are on GPU cloud providers compared.
Every price in the index
The plan as the provider lists it, its price per GPU-hour and the smallest number of GPUs you can rent at that price. Prices are rounded to the cent here; the CSV keeps each provider's exact figure.
| Provider | Plan as listed | Per GPU-hour | Smallest size | Checked |
|---|---|---|---|---|
| B200 · median $6.99 | ||||
| Hyperstack | B200 192GB | $6.00 | 8 GPUs | Oct 18, 2026 |
 Lambda | B200 180GB SXM6 (1x instance) | $6.99 | 1 GPU | Oct 17, 2026 |
| RunPod | B200 180GB (Secure Cloud) | $7.99 | 1 GPU | Oct 18, 2026 |
| H200 · median $4.47 | ||||
| Hyperstack | H200 141GB SXM | $3.99 | 8 GPUs | Oct 18, 2026 |
 DigitalOcean GPU Droplets | HGX H200 141GB | $4.47 | 1 GPU | Oct 17, 2026 |
| RunPod | H200 141GB (Secure Cloud) | $5.29 | 1 GPU | Oct 18, 2026 |
| H100 SXM · median $4.14 | ||||
| Hyperstack | H100 80GB SXM | $3.20 | 8 GPUs | Oct 18, 2026 |
| RunPod | H100 80GB SXM (Secure Cloud) | $3.99 | 1 GPU | Oct 18, 2026 |
| Lambda | H100 80GB SXM (1x instance) | $4.29 | 1 GPU | Oct 17, 2026 |
| DigitalOcean GPU Droplets | HGX H100 80GB | $4.41 | 1 GPU | Oct 17, 2026 |
| A100 80GB PCIe · median $1.79 | ||||
| Hyperstack | A100 80GB (PCIe) | $1.35 | 1 GPU | Oct 18, 2026 |
| RunPod | A100 80GB PCIe (Secure Cloud) | $1.79 | 1 GPU | Oct 18, 2026 |
 Vultr Cloud GPU | A100 80GB PCIe (1 GPU) | $2.40 | 1 GPU | Oct 17, 2026 |
| L40S · median $1.57 | ||||
| RunPod | L40S 48GB (Secure Cloud) | $1.09 | 1 GPU | Oct 18, 2026 |
| DigitalOcean GPU Droplets | L40S 48GB | $1.57 | 1 GPU | Oct 17, 2026 |
| Vultr Cloud GPU | L40S 48GB (1 GPU) | $1.67 | 1 GPU | Oct 17, 2026 |
How we collect this
- Every price is the on-demand rate on the provider's own public pricing page, in US dollars per GPU per hour, before tax. We record the date we read it; storage, data transfer and tax can add to the bill.
- One price per provider and GPU: the smallest size it rents on demand, usually one GPU. Bigger machines are often a little cheaper per GPU, but you have to rent them whole. Where a provider sells a GPU only in an 8-GPU machine, the table says so, and "cheapest single GPU" gives the lowest price you can rent as one GPU.
- SXM and PCIe versions of a GPU have separate rows, because they differ in speed and price. HGX servers use SXM modules, so they count as SXM.
- The median is the middle price, or the average of the two middle prices when the count is even. A GPU needs prices from at least 3 providers to get a row.
- The cheapest price is not always the best buy: providers differ in vCPUs, RAM, storage, regions and how often a GPU is in stock. Check availability before you plan around a price.
What's left out, and why
- TensorDock: Shows only "from" prices without a login, so we can't check a real on-demand price.
- Vast.ai: A marketplace of independent hosts, and its price data may not be used in an index without a data licence.
- RunPod Community Cloud: GPUs in third-party data centres, closer to a marketplace than a cloud.
- Serverless GPUs (Modal): Billed per second for the GPU alone, with CPU and memory on top, so not comparable with a rented machine.
- Spot prices (Hyperstack): The provider can stop a spot machine at any time.
- Vultr Cloud GPU HGX H100 80GB SXM (8-GPU node): A "starting at" price for a whole 8-GPU server; Vultr's own API listed the server as not deployable on demand.
- Vultr Cloud GPU HGX B200 192GB (8-GPU node): A "starting at" price for a whole 8-GPU server; Vultr's own API listed the server as not deployable on demand.
This is the first edition. When we re-check a price, this page, its date and the CSV files change with it. More on our method in how we work and our editorial policy. Seen a price that has changed? Email hello@askgpu.com.
Cite this page
You're welcome to quote these numbers or use the CSV files. Please credit AskGPU and link to this page, with the date, since GPU prices change often.
AskGPU, "GPU rental price index", October 2026 edition, prices checked Oct 17–18, 2026. https://askgpu.com/gpu-price-index/
Downloads: the index (one row per GPU) and every price (one row per provider and GPU), as CSV.