What it is
Runpod rents GPUs three ways from one account. Pods are GPU machines you rent by the hour, either in Secure Cloud (vetted data centres) or the cheaper Community Cloud (third-party hosts). Serverless runs your container as workers that scale to zero when idle and are billed by the second. Clusters provide multi-GPU machines linked by InfiniBand for training and large batch jobs.
According to the pricing page, an H100 SXM pod costs $3.99 an hour in Secure Cloud and $2.69 in Community Cloud, and an A100 SXM costs $1.79 and $1.39. Serverless workers are priced higher, about $4.79 an hour for an H100 and $2.72 for an A100 80GB, because you pay only while a worker runs. Runpod says it operates in 31 regions with more than 30 GPU types.
Our review
Runpod's case rests on price. According to its pricing page, an H100 SXM pod costs $3.99 an hour in Secure Cloud and $2.69 in Community Cloud, an RTX 4090 just $0.34 on Community Cloud, and an A100 SXM $1.39. For long training runs, fine-tuning and experiments, few managed platforms charge less for an on-demand hour.
Serverless is the part that competes with Modal and Baseten. You hand Runpod a container and it runs workers that scale from zero, billed per second and rounded up. The docs are frank that cold start, loading the model into GPU memory, is billed, and that an idle worker stays on for 5 seconds by default before scaling down. Active workers that never sleep are discounted only through sales, and the homepage's sub-200 ms FlashBoot cold starts are the vendor's own figure. New accounts also hit a default spend cap of $80 an hour until support raises it.
The two-tier cloud is the trap to understand. Runpod's compliance page lists a SOC 2 Type II report, a SOC 3 and an ISO 27001:2022 certificate issued in August 2026, and describes HIPAA and GDPR as programmes rather than certifications. It also says the ISO certificate covers the management system, not individual products, and that coverage varies by region and deployment model. The page does not say whether Community Cloud pods on third-party hosts are covered, and Runpod's own security guide says its documentation recommends single-tenant Secure Cloud for sensitive workloads. Anything sensitive belongs on Secure Cloud, at roughly 50% more per H100 hour.
The developer experience is plainer than Modal's. You build and push Docker images rather than decorating Python functions, and there is no shared per-token API for popular models to fall back on. Storage is easy to forget: a stopped pod's volume disk is billed at $0.20 per GB a month, double the running rate, so a fleet of parked 500 GB pods quietly costs $100 each per month.
Where it earns its place
An ML engineer fine-tuning a 70B model over a weekend
Rent eight H100s for 48 hours without a contract
Eight H100 SXM pods on Secure Cloud cost $31.92 an hour, about $1,530 for 48 hours; on Community Cloud the same run is about $1,030. Network storage for checkpoints adds $0.07 per GB a month.
A two-person start-up serving an image model with spiky traffic
Run an inference endpoint that costs nothing overnight
Package the model in a container and deploy it on Serverless flex workers, billed by the second at about $1.10 an hour on an RTX 4090. Cold starts are billed, so cache the model in the image or use FlashBoot to keep them short.
A security lead at a health-data company reviewing GPU vendors
Confirm a GPU cloud fits HIPAA and SOC 2 requirements
Request the SOC 2 Type II report and ISO certificate from the Runpod Trust Center and confirm in writing that workloads run on Secure Cloud. The compliance page does not mention a BAA, so ask for one during procurement.
Before you commit
- Setup
- Add credit, pick a GPU and a template such as PyTorch, and a pod starts in minutes; Serverless needs a Docker image with a handler.
- Integrations
- Web console, REST API, CLI and Python SDK; containers from any registry.
- Data and privacy
- SOC 2 Type II, SOC 3 and ISO 27001:2022; HIPAA and GDPR programmes with coverage that varies by region and deployment model.
- Cost at scale
- A single H100 SXM on Secure Cloud running all month costs about $2,900 at $3.99 an hour; reserved clusters of 1 to 12 months are priced by sales.
- Limits
- Default account spend cap of $80 an hour; serverless idle timeout defaults to 5 seconds; Community Cloud availability depends on third-party hosts.
What's good
- Among the lowest on-demand GPU prices, H100 SXM from $2.69 an hour
- Pods, serverless and multi-node clusters in one account
- Serverless billed per second with scale to zero
- SOC 2 Type II, SOC 3 and ISO 27001:2022
What's not
- Community Cloud sits outside the clearest compliance coverage
- Active-worker and reserved discounts only through sales
- Cold-start time is billed
- Idle pod volumes cost double the running rate
- You build and maintain your own containers
Best for
ML engineers who want cheap GPUs for training, fine-tuning or a self-managed inference endpoint and can work with Docker.
Alternatives
Runpod: common questions
- Is Runpod worth it?
- Runpod is where cheap GPU hours live, provided you are happy to package your own containers and to check which cloud tier your compliance actually covers. We rate it 4.3 out of 5.
- How much does Runpod cost?
- Pods per hour: H100 SXM $3.99 Secure / $2.69 Community; A100 SXM $1.79 / $1.39; B200 $7.99 / $5.98. Serverless per second: H100 about $4.79/hour, A100 80GB $2.72/hour, 4090 $1.10/hour. Storage from $0.05/GB/month.
- What are the downsides of Runpod?
- Community Cloud sits outside the clearest compliance coverage; active-worker and reserved discounts only through sales; cold-start time is billed; idle pod volumes cost double the running rate; you build and maintain your own containers.
- Who is Runpod best for?
- ML engineers who want cheap GPUs for training, fine-tuning or a self-managed inference endpoint and can work with Docker.
- How much does Runpod cost for a team or in production?
- A single H100 SXM on Secure Cloud running all month costs about $2,900 at $3.99 an hour; reserved clusters of 1 to 12 months are priced by sales.
- Is Runpod safe for company data?
- SOC 2 Type II, SOC 3 and ISO 27001:2022; HIPAA and GDPR programmes with coverage that varies by region and deployment model.
- What are the best Runpod alternatives?
- The closest alternatives we have reviewed are Modal, Baseten, Replicate and Together AI.
Answers checked Oct 23, 2026.
Make Runpod? Show this rating on your site
