RunPod pricing: GPU pods, storage fees and real bills

RunPod charges by the second for GPU pods, from $0.34 an hour for an RTX 4090 on its Community Cloud to $8.99 an hour for a B300 on its Secure Cloud, as of our October 18, 2026 price check. Storage is billed on top, from $0.05 to $0.20 per GB per month, and there are no fees for data in or out. One H100 SXM running for a 720-hour month costs $1,936.80 on Community Cloud or $2,872.80 on Secure Cloud, before storage and any sales tax.

We haven’t rented RunPod’s machines ourselves. GPU prices come from the site’s own price data for RunPod (checked October 18, 2026, when RunPod’s pricing page said “Updated October 8, 2026”). Billing rules, storage rates and limits come from RunPod’s documentation, read on November 7, 2026. Prices for other providers are from our data files, with their dates given below. RunPod’s console shows the live price when you deploy, so check it before you launch.

How much does a RunPod GPU cost per hour?

RunPod sells GPU “pods” in two clouds. RunPod’s docs describe Secure Cloud as running “in T3/T4 data centers, providing high reliability and security for enterprise and production workloads.” Community Cloud “connects individual compute providers to users through a vetted, secure peer-to-peer system” at lower prices. Here is the on-demand rate card, per GPU per hour.

GPU (memory)	Secure Cloud	Community Cloud	Community saving
RTX 4090 (24 GB)	$0.89	$0.34	62%
RTX 5090 (32 GB)	$1.19	$0.69	42%
L40S (48 GB)	$1.09	$0.79	28%
A100 PCIe (80 GB)	$1.79	$1.19	34%
A100 SXM (80 GB)	$1.79	$1.39	22%
H100 PCIe (80 GB)	$2.89	$1.99	31%
H100 SXM (80 GB)	$3.99	$2.69	33%
H100 NVL (94 GB)	$3.19	$2.59	19%
H200 (141 GB)	$5.29	$3.59	32%
B200 (180 GB)	$7.99	$5.98	25%
B300 (288 GB)	$8.99	$6.94	23%

Prices in US dollars per GPU per hour, on demand, from our RunPod data checked October 18, 2026. The saving column is our arithmetic: (Secure − Community) ÷ Secure, rounded.

A few things stand out:

• The cheapest cards have the biggest gap. An RTX 4090 costs 62% less on Community Cloud. For the H100 SXM the gap is 33%, and for the B200 and B300 it’s about a quarter.
• The A100 SXM and A100 PCIe cost the same on Secure Cloud ($1.79). Our data shows the SXM version comes with 16 vCPUs and 125 GB of RAM, against 8 vCPUs and 117 GB for the PCIe pod.
• The L40S on Secure Cloud ($1.09) costs less than the RTX 5090 ($1.19) and has 48 GB of memory instead of 32 GB.
• CPU and RAM come with the GPU. On Secure Cloud our data lists, for example, 6 vCPUs and 41 GB of RAM with an RTX 4090 and 20 vCPUs and 125 GB with an H100 SXM.

Is there a free tier or a discount for committing?

There’s no free tier. Our data records “no free credits” and only a small referral bonus once a new user loads $10. RunPod’s billing docs say new users “can start with as little as $10.”

There’s also no longer a way to buy a discount for steady use. RunPod’s pod pricing docs say savings plans are deprecated and can’t be bought any more. Plans bought earlier stay active until they expire, cover only GPU compute for pods of the same GPU type, and are non-refundable. For large or enterprise workloads, RunPod says custom plans are available through its sales team, without published prices.

What does RunPod storage cost?

Every pod has a container disk and usually a volume disk, and you can add network or global volumes. Each is billed differently. These rates are from RunPod’s pod pricing, billing and storage docs, read November 7, 2026.

Storage	While pod runs	While pod is stopped	What happens to the data
Container disk	$0.10/GB/month	Not charged	Erased when the pod stops
Volume disk	$0.10/GB/month	$0.20/GB/month	Kept until the pod is terminated
Network volume, first 1 TB	$0.07/GB/month	$0.07/GB/month	Kept after the pod is deleted
Network volume, beyond 1 TB	$0.05/GB/month	$0.05/GB/month	Same
Global volume (beta)	$0.09/GB/month plus IOPS	Same	Kept after the pod is stopped or deleted

Container and volume disks bill per second; network volumes bill hourly. RunPod says you aren’t charged for storage while the host machine is unavailable, and that RunPod “is not designed for long-term storage,” so back up anything critical elsewhere.

What the table means in dollars:

• A 200 GB volume disk costs 200 × $0.10 = $20 a month while the pod runs, and 200 × $0.20 = $40 a month while it sits stopped. Stopping a pod doesn’t stop this charge.
• A 200 GB network volume costs 200 × $0.07 = $14 a month, running or not.
• A 500 GB network volume costs 500 × $0.07 = $35 a month.

So if you keep model weights between sessions, a network volume costs less than a stopped volume disk of the same size: $14 against $40 for 200 GB.

Network volume rules that affect the bill

From RunPod’s network volume docs:

• Size can go up, not down. “Volume size can be increased later but cannot be decreased.” Volumes over 4 TB need a request to support.
• Pods: Secure Cloud only. On pods, network volumes are available only on Secure Cloud. A Community Cloud pod keeps its data on its own volume disk.
• Attach at launch. A network volume must be attached when the pod is deployed and can’t be attached or detached later without deleting the pod.
• Tied to one data center. Each volume is created in a specific data center. A Serverless endpoint with one volume can only run workers in that data center.
• A faster tier exists. RunPod lists a High-Performance tier with “up to 3x throughput and 4x IOPS” in select data centers, priced at a premium. The docs we read don’t give its price.
• An S3-compatible API lets you manage files on a volume without starting a pod.

What are RunPod global volumes?

Global volumes are a newer, beta storage type that isn’t tied to one data center. RunPod’s storage options page lists them at $0.09 per GB per month plus IOPS. The global volume overview adds request charges: $0.005 per 1,000 Class A requests (writes, creates, lists) and $0.0005 per 1,000 Class B requests (reads, metadata lookups). Storage grows with your data, so there’s no size to choose.

The limits matter more than the price. RunPod says global volumes are not fully POSIX compliant (no file locking, atomic rename or hard links), that when two pods write the same file “the last write wins,” and that updates may not show up immediately. A pod can have one global volume and one network volume at a time. If your balance reaches $0, a global volume is “flagged and permanently deleted after 15 days.” RunPod suggests a network volume for fast writes or several pods writing the same files.

Global networking is a separate feature: a private network between pods. RunPod’s networking docs list it for NVIDIA GPU pods in 17 data centers, running at 100 Mbps between pods. The page doesn’t say whether it costs extra.

How does RunPod billing work?

RunPod is prepaid. You buy credits, and pods and storage draw them down by the second. The rules below are from RunPod’s billing and pod pricing docs.

• Payment: cards through Stripe (prepaid cards must deposit at least $100 per transaction), cryptocurrency (with KYC checks), or business invoicing by ACH, wire or card for transactions over $5,000, set up through sales.
• To launch a pod, your balance must cover at least one hour of the configuration you pick. For one H100 SXM on Secure Cloud that is $3.99; for eight it is 8 × $3.99 = $31.92.
• Spend limit: the default is $80 an hour across all resources. RunPod says it “increases over time based on account history,” or you can ask support. At $3.99 each, that is 20 H100 SXM GPUs on Secure Cloud (20 × $3.99 = $79.80) before storage.
• Credits are non-refundable and can’t be withdrawn. They can only be spent on RunPod. The billing page doesn’t mention an expiry date.
• Sales tax is added “on top of” credit purchases in some jurisdictions, based on your billing address and payment method. RunPod’s example: at an 18% rate, $100 of credit costs $118. A valid tax ID may exempt a business.
• Auto-pay reloads your balance from a saved card when it falls below a threshold you set, at most once an hour. Low-balance emails are separate; you can use either or both.

What happens when your balance hits $0?

This is the rule that costs people data. When the balance reaches $0, RunPod stops all running pods:

• Pods with a network volume are stopped and the volume’s data is kept, but storage keeps billing. If the charges stay unpaid, RunPod says the volume “may eventually be terminated” and the data can’t be recovered.
• Pods without a network volume are terminated, and their data “can’t be recovered.”

Since Community Cloud pods can’t use network volumes, a Community pod that runs out of credit loses its disk. Turn on auto-pay or a low-balance email before a long job.

What about RunPod Serverless?

RunPod Serverless runs your container as workers behind an endpoint. Its docs describe two worker types: flex workers scale to zero when there are no requests and bill at the standard per-second rate, and active workers run around the clock, with discounts available through sales.

Serverless rates per GPU are listed on RunPod’s pricing page. They aren’t in our price data, so this guide doesn’t quote them. What RunPod’s serverless pricing docs do say about billing:

• Billing runs per second, from when a worker starts until it fully stops, rounded up to the nearest second.
• Cold starts are billed from worker start until your handler is ready, which is mostly loading the model into GPU memory. Pulling the container image and downloading cached models while scaled down are not billed.
• After its last job, a worker waits for new requests for the idle timeout, 5 seconds by default, and that wait is billed.
• Container disk costs about $0.10/GB/month, billed in 5-minute intervals. Network volumes cost the same as for pods.

For how per-second serverless prices compare at 1, 10 and 100 hours of use, see our guide to serverless GPU pricing on Modal, Replicate and Together AI.

What does RunPod cost per month? Three worked examples

1. A hobbyist running Stable Diffusion or a small model, 3 hours a day. 30 days × 3 hours = 90 hours.

• Community Cloud RTX 4090: 90 × $0.34 = $30.60. Add a 50 GB volume disk. It is stopped most of the month, so budget up to 50 × $0.20 = $10. About $40.60 a month.
• Secure Cloud RTX 4090: 90 × $0.89 = $80.10, plus a 50 GB network volume at 50 × $0.07 = $3.50. $83.60 a month.

Community Cloud is about half the price here. Secure Cloud buys the network volume, which survives a $0 balance and can be reused by the next pod.

2. A weekend of fine-tuning on one H100 SXM. 48 hours of compute plus a 200 GB network volume kept for the month.

• Secure Cloud: 48 × $3.99 = $191.52, plus $14 of storage = $205.52.
• Community Cloud: 48 × $2.69 = $129.12. There’s no network volume on Community pods, so use a 200 GB volume disk: about 48 hours at $0.10 and the rest of the month at $0.20, roughly $39. About $168, or $129.12 plus a few dollars if you terminate the pod and keep your files elsewhere.

3. An always-on inference server on one 80 GB GPU. 720 hours.

GPU	Secure Cloud	Community Cloud
A100 PCIe 80 GB	720 × $1.79 = $1,288.80	720 × $1.19 = $856.80
H100 PCIe 80 GB	720 × $2.89 = $2,080.80	720 × $1.99 = $1,432.80
H100 SXM 80 GB	720 × $3.99 = $2,872.80	720 × $2.69 = $1,936.80

Over a year, an H100 SXM on Secure Cloud is 12 × $2,872.80 = $34,473.60 before storage and tax. If traffic is uneven, a serverless endpoint that scales to zero can cost less, because idle time isn’t billed after the idle timeout.

Is RunPod cheaper than Vast.ai, Lambda and the rest?

Sometimes. RunPod Community Cloud is among the cheapest fixed prices for big data-center GPUs, while marketplaces undercut it on consumer cards. Here are the same GPUs at eight providers, per GPU per hour, on demand.

GPU	RunPod Secure / Community	Vast.ai median	Lambda	Hyperstack	Others
RTX 4090	$0.89 / $0.34	$0.47	–	–	TensorDock from $0.35
A100 80 GB	$1.79 / $1.19 (PCIe), $1.39 (SXM)	$0.80 (SXM4)	$2.79 (SXM, 8x only)	$1.35 (PCIe)	Modal $2.4984; Vultr $2.397 (PCIe)
H100 SXM	$3.99 / $2.69	$2.40	$4.29 (1x), $3.99 (8x)	$3.20 (8-GPU VM only)	DigitalOcean $4.41; Modal $3.9492
H200	$5.29 / $3.59	–	–	$3.99 (8-GPU VM only)	DigitalOcean $4.47; Modal $4.5396
B200	$7.99 / $5.98	$8.63 (192 GB)	$6.99 (1x), $6.69 (8x)	$6.00 (8-GPU VM only)	Modal $6.2496

Sources: our price data for RunPod (October 18, 2026), Vast.ai (median offers, November 2), Lambda (November 2), Hyperstack (October 18), DigitalOcean GPU Droplets (November 2), Modal (November 2; GPU only, CPU and memory billed extra), Vultr (October 17) and TensorDock (October 17). TensorDock’s figures are unverified homepage “from” prices. Vultr also lists an H100 at $2.30 and a B200 at $3.50 per GPU, but only as 8-GPU servers whose “starting from” prices its bare-metal section ties to prepaid contracts, so we’ve left them out of the table.

Worked example: one H100 SXM for a 720-hour month.

• Vast.ai median host: 720 × $2.40 = $1,728.00
• RunPod Community Cloud: 720 × $2.69 = $1,936.80
• RunPod Secure Cloud: 720 × $3.99 = $2,872.80
• Lambda (1 GPU): 720 × $4.29 = $3,088.80
• DigitalOcean: 720 × $4.41 = $3,175.20

RunPod Community Cloud costs $208.80 a month more than Vast.ai’s median H100 host and $1,152 less than Lambda’s single H100. For a B200, RunPod Community’s $5.98 is the lowest single-GPU price in our data; Hyperstack’s $6.00 needs an 8-GPU VM.

The hourly rate isn’t the whole bill: RunPod’s docs say it charges no fees for data in or out, and storage is billed on top. For a full side-by-side, see RunPod vs Vast.ai and the Lambda vs RunPod comparison; for the cheapest place to rent each GPU, see the cheapest cloud GPU guide.

What to watch for

• Stopped is not free. A stopped pod’s volume disk bills at $0.20/GB a month, double the running rate. Terminate pods you don’t need, after moving the data you want to keep.
• $0 balance can delete your work. Pods without a network volume are terminated when credit runs out. Set up auto-pay or a low-balance email.
• Community pods can’t use network volumes. If you need storage that survives a terminated pod, use Secure Cloud or copy data out.
• A stopped pod may come back without a GPU. RunPod’s docs say starting a stopped pod can allocate zero GPUs if capacity has changed.
• Editing a running pod resets it and erases everything outside /workspace.
• Network volumes only grow. You can’t shrink one, so size it carefully.
• No refunds. Credits are non-refundable and can’t be withdrawn, so top up in amounts you expect to use.
• The $80-an-hour spend limit caps how many GPUs a new account can run at once until RunPod raises it.
• Tax may be added at checkout, depending on where you are.
• No commitment discount for new buyers, since savings plans are deprecated.

Which RunPod option should you choose?

• Learning, image generation or small models on a budget: a Community Cloud RTX 4090 at $0.34 an hour. Keep the disk small and terminate pods you aren’t using.
• Fine-tuning for a few days: an H100 SXM on Community Cloud ($2.69) if you can afford to lose the pod’s disk, or Secure Cloud ($3.99) with a network volume if you can’t.
• A production API with steady traffic: a Secure Cloud pod, sized to the smallest 80 GB GPU your model runs on. The H100 PCIe ($2.89) costs $792 less a month than the H100 SXM ($3.99) at 720 hours.
• A production API with uneven traffic: RunPod Serverless with flex workers, checking the current per-second rate on RunPod’s pricing page.
• The lowest price for an H100 or A100: compare Vast.ai’s marketplace first, with the reliability trade-offs set out in RunPod vs Vast.ai.
• You want fixed prices from a single data-center company: see the options in our RunPod alternatives guide and on the GPU cloud hub.

Our RunPod page keeps the current rate card from our data.

Sources

• RunPod Docs: Pod pricing, read November 7, 2026
• RunPod Docs: Billing overview, read November 7, 2026
• RunPod Docs: Serverless pricing, read November 7, 2026
• RunPod Docs: Storage options, read November 7, 2026
• RunPod Docs: Network volumes, read November 7, 2026
• RunPod Docs: Global volumes overview, read November 7, 2026
• RunPod Docs: Global networking, read November 7, 2026
• RunPod Docs: Pods overview, read November 7, 2026
• RunPod Docs: Manage Pods, read November 7, 2026
• RunPod Docs: Sales tax on Runpod purchases, read November 7, 2026
• AskGPU price data for RunPod, Vast.ai, Lambda, Hyperstack, DigitalOcean GPU Droplets, Modal, Vultr and TensorDock (src/data/products.json), checked October 17 to November 2, 2026