Before you compare hourly prices, check four things. How finely are you billed, what does a stopped machine cost, where does it run, and does the provider say?
This page compares Runpod, Novita AI, DigitalOcean and UpCloud, whose prices we check every week, and Vast.ai's marketplace. The facts come from each provider's pricing and documentation pages, read on 27 October 2026.
Who this is for
- you need a GPU for hours or days, not a whole month
- you want to compare per-second billing, storage charges and locations
- you are choosing between Runpod and Vast.ai
Not for you if you need the GPU all month: see dedicated GPU servers by the month.
Key numbers
- 672 hours a month, after which UpCloud stops charging for an hourly GPU server Official page,
- 60 GB of container disk included with a Novita GPU instance before disk charges start Official page,
- 80% the most Vast.ai says its interruptible rentals save against on-demand ones Official page,
The GPU clouds side by side
Most bill by the second; what a stopped machine costs differs the most.
| Billing | GPUs in our records | Where | While stopped | |
|---|---|---|---|---|
| Runpod | By the second, compute and storage | RTX 4090, RTX 5090, L40S, A100, H100, H200 | 31 regions; no page says which GPU is where | Volume disk still billed; container disk not |
| Novita AI | To the second, settled every hour | RTX 4090, RTX 5090, L40S | 14 regions, not named | Compute stops; disks priced per GB per day |
| DigitalOcean | By the second | L40S, H100, a server with eight H100s | H100: New York, Amsterdam, Toronto; L40S: Toronto | Billed until you destroy the GPU Droplet |
| UpCloud | By the started hour, up to 672 hours a month | L4, L40S, H100 (one, two or eight) | Helsinki | Not stated for GPU servers |
| Vast.ai | By the second | Hosts' own machines, in a marketplace | Set by each host | Storage billed until you delete the instance |
Runpod and Novita price each GPU on its own. DigitalOcean and UpCloud price the whole server, with its processor cores and memory.
Runpod's prices below are for its Secure Cloud, in Tier III and Tier IV data centres run by vetted partners. Its Community Cloud charges less per hour for the same GPUs.
Prices by card
24 and 32 GB cards
These cards run models of about 7 to 14 billion parameters, by our GPU memory guide.
| Plan | Price | New-customer offer | Link |
|---|---|---|---|
| Runpod RTX 4090 24 GB GPU | $0.89/hour (about $650/month) Secure Cloud, on demand checked 27 Oct 2026 | No offer | Visit |
| Novita AI GPU instance, RTX 4090 24GB 24 GB GPU | $0.33/hour (about $241/month) On demand, per GPU; billed by the second, storage extra checked 27 Oct 2026 | No offer | Visit |
| Runpod RTX 5090 32 GB GPU | $1.19/hour (about $869/month) Secure Cloud, on demand checked 27 Oct 2026 | No offer | Visit |
| Novita AI GPU instance, RTX 5090 32GB 32 GB GPU | $0.56/hour (about $409/month) On demand, per GPU; billed by the second, storage extra checked 27 Oct 2026 | No offer | Visit |
| UpCloud GPU Server, 1 x NVIDIA L4, 8 cores, 64 GB 24 GB GPU | €0.58/hour (about €390/month; no charge after 28 days of use)30-day month: the last 2 days are free.31-day month: the last 3 days are free.February: no free days (one in a leap year).Started mid-month: it may never reach the cap that month. on demand, billed by the starting hour; monthly cost capped at 28 days of use checked 25 Oct 2026 | No offer | Visit |
Monthly costs for hourly prices assume the server runs all month (730 hours), or less where a provider caps the monthly bill.
Runpod and Novita price one GPU; UpCloud's price is for the whole server, in euros.
48 GB cards
| Plan | Price | New-customer offer | Link |
|---|---|---|---|
| Runpod L40S 48 GB GPU | $1.09/hour (about $796/month) Secure Cloud, on demand checked 27 Oct 2026 | No offer | Visit |
| Novita AI GPU instance, NVIDIA L40S 48GB 48 GB GPU | $0.55/hour (about $402/month) On demand, per GPU; billed by the second, storage extra checked 27 Oct 2026 | No offer | Visit |
| DigitalOcean GPU Droplet, 1 x NVIDIA L40S 48 GB GPU | $1.57/hour (about $1,146/month) on demand, billed per second · VAT not stated checked 25 Oct 2026 | New accounts get a $5 credit that lasts 90 days and covers every product except SaaS add-ons checked 29 Oct 2026 | Visit |
| UpCloud GPU Server, 1 x NVIDIA L40S, 8 cores, 64 GB 48 GB GPU | €1.11/hour (about €746/month; no charge after 28 days of use)30-day month: the last 2 days are free.31-day month: the last 3 days are free.February: no free days (one in a leap year).Started mid-month: it may never reach the cap that month. on demand, billed by the starting hour; monthly cost capped at 28 days of use checked 25 Oct 2026 | No offer | Visit |
Hourly plans show the price for a full month as well; you pay only for the hours you use.
80 GB cards and larger
| Plan | Price | New-customer offer | Link |
|---|---|---|---|
| Runpod A100 PCIe 80 GB GPU | $1.79/hour (about $1,307/month) Secure Cloud, on demand checked 27 Oct 2026 | No offer | Visit |
| Runpod H100 PCIe 80 GB GPU | $2.89/hour (about $2,110/month) Secure Cloud, on demand checked 27 Oct 2026 | No offer | Visit |
| Runpod H100 SXM 80 GB GPU | $3.99/hour (about $2,913/month) Secure Cloud, on demand checked 27 Oct 2026 | No offer | Visit |
| Runpod H200 141 GB GPU | $5.29/hour (about $3,862/month) Secure Cloud, on demand checked 27 Oct 2026 | No offer | Visit |
| DigitalOcean GPU Droplet, 1 x NVIDIA H100 80 GB GPU | $4.41/hour (about $3,219/month) on demand, billed per second · VAT not stated checked 25 Oct 2026 | New accounts get a $5 credit that lasts 90 days and covers every product except SaaS add-ons checked 29 Oct 2026 | Visit |
| UpCloud GPU Server, 1 x NVIDIA H100, 12 cores, 240 GB 80 GB GPU | €2.17/hour (about €1,458/month; no charge after 28 days of use)30-day month: the last 2 days are free.31-day month: the last 3 days are free.February: no free days (one in a leap year).Started mid-month: it may never reach the cap that month. on demand, billed by the starting hour; monthly cost capped at 28 days of use checked 25 Oct 2026 | No offer | Visit |
An 80 GB card runs a 70B model at 4-bit; the H200 has 141 GB.
| Plan | Price | New-customer offer | Link |
|---|---|---|---|
| DigitalOcean GPU Droplet, 8 x NVIDIA H100 8× 80 GB GPUs | $35.28/hour (about $25,754/month) on demand, billed per second; price for the whole 8-GPU machine · VAT not stated checked 25 Oct 2026 | New accounts get a $5 credit that lasts 90 days and covers every product except SaaS add-ons checked 29 Oct 2026 | Visit |
| UpCloud GPU Server, 8 x NVIDIA H100, 96 cores, 1920 GB 8× 80 GB GPUs | €17.36/hour (about €11,666/month; no charge after 28 days of use)30-day month: the last 2 days are free.31-day month: the last 3 days are free.February: no free days (one in a leap year).Started mid-month: it may never reach the cap that month. on demand, billed by the starting hour; price for the whole 8-GPU server checked 25 Oct 2026 | No offer | Visit |
Both prices are for the whole machine; UpCloud's is in euros.
Our benchmark page shows how fast each card runs a model, from published results.
Picks by need
- Data in the EU: UpCloud in Helsinki, or DigitalOcean's H100 in Amsterdam.
- Paying only while it runs: Runpod charges only for disks while a pod is stopped, and Novita stops compute billing when you stop the instance. DigitalOcean bills a GPU Droplet until you destroy it.
- A whole eight-GPU server with a stated price: DigitalOcean and UpCloud. Runpod prices single GPUs.
- Lower rates, if interruptions are fine: Vast.ai's interruptible rentals or Runpod's Community Cloud.
- No machine at all: pay per token instead, as our guide to open models without a server explains.
Runpod or Vast.ai
Runpod rents pods on its Secure Cloud or on a cheaper Community Cloud. It bills compute and storage by the second, and needs enough credit for one hour of the chosen pod before it starts one.
Vast.ai is a marketplace where hosts set the prices. On-demand rentals cannot be interrupted. Interruptible ones run on your bid, and a higher bid stops them, killing the running processes.
Vast.ai's docs say higher reliability scores usually come with higher prices. Its storage is billed for as long as the instance exists, even while it is stopped.
Choose Runpod for a predictable machine in a vetted data centre. Choose Vast.ai for jobs that can restart after an interruption, at marketplace prices that change through the day.
Runpod's provider page lists all its plans with their checked prices.
What to look for
- Billing steps: by the second, or by each started hour. Short tests cost less with per-second billing.
- Storage while stopped: a stopped machine can still cost money until you delete it or its disks.
- Where it runs: if your data must stay in the EU, pick a provider that names the city for your GPU.
- Who owns the machines: Runpod's Secure Cloud uses partners' data centres; Vast.ai's hosts are third parties.
- Security papers: Runpod lists SOC 2 (Type II) and ISO/IEC 27001:2022 on its own site.
A quick safety check before you rent
- Find the company's legal name and country in its terms.
- Read how it bills stopped machines and storage.
- Start with a small top-up; Runpod stops pods when the balance runs out.
- Keep your files somewhere you control, and download results before you delete a machine.
Our guide to stopping a runaway AI bill covers spending alerts and prepaid credit.
How we checked
We read Runpod's, Novita AI's, DigitalOcean's, UpCloud's and Vast.ai's own pricing and documentation pages on 27 October 2026. Prices come from our weekly price records; Vast.ai's marketplace prices change every hour and are not in them.
Our own measurements will be added here, with their dates, when our test bench runs them.
Mistakes to avoid
Powering off instead of deleting
Instead: DigitalOcean bills powered-off GPU Droplets; Vast.ai bills storage until you delete the instance.
Keeping work on a disk that is wiped
Instead: Runpod erases the container disk when a pod stops; keep files on a volume.
Running a long job on an interruptible rental without saving
Instead: A higher bid can stop it at any time; save checkpoints often.
Assuming EU hosting
Instead: Pick a provider that names the city for your GPU, such as UpCloud or DigitalOcean.
Other questions
Which GPU cloud is cheapest by the hour?
It changes from week to week; the boxes above show the prices we checked most recently. Vast.ai says its interruptible rentals cost 50 to 80% less than its on-demand ones, but they stop when someone bids higher.
Is Runpod or Vast.ai better?
Runpod suits a predictable on-demand machine in a vetted data centre. Vast.ai suits jobs that can survive an interruption, at marketplace prices.
Do GPU clouds charge when the machine is off?
Often, yes. DigitalOcean bills a powered-off GPU Droplet, Runpod bills a stopped pod's volume disk, and Vast.ai bills storage until you delete the instance.
Sources and test notes
- Runpod: pricing, read
- Runpod documentation: Pod pricing (billing, stopped Pods, credit), read
- Novita AI: GPU instances and pricing table, read
- Novita AI documentation: GPU instance pricing, read
- DigitalOcean documentation: Droplet pricing and billing, read
- UpCloud: pricing, read
- UpCloud documentation: getting started FAQ (hourly billing and monthly cap), read
- Vast.ai documentation: pricing (billing, storage, reliability), read
- Vast.ai documentation: rental types FAQ, read
Every page read on 27 October 2026. Prices come from our weekly price records; each box shows its check date.