Runpod
Runpod provides GPU Pods (per-hour instances), Serverless autoscaling inference workers, multi-node Clusters, persistent storage, and a public model endpoint marketplace with per-request or per-token pricing. Documentation shows many GPU SKUs across 31 global regions and a mix of per-hour and per-request prices for compute and model endpoints.
The pricing pages list per-hour Pod GPU rates (examples include H100 at $2.89/hr), Serverless worker hourly rates, Cluster hourly rates for some SKUs and 'Contact sales' for others, storage per-GB monthly rates, and per-request or per-token prices for public endpoints.Is Runpod right for you?
Good for
- AI teams that need flexible access to many GPU SKUs and regions for experiment-to-production work, with per-hour instance and per-request model pricing.
- Wide global footprint and many GPU SKUs let you place workloads in regions to reduce latency.
- Serverless endpoints claim scale-to-zero behavior with very low cold-start latency, useful for traffic that bursts from zero.
Keep in mind
- Reserved cluster and long-term discounted pricing are not published on the public pages and require contacting sales.
- Compliance coverage and guarantees can differ by region and deployment model, so you must verify region-specific compliance availability.
- Buyers who expect published long-term reserved cluster discounts and multi-month contract terms on public pricing pages—those require contacting sales for details.
Pricing and plan limits
The linked official pages show many hourly and per-request prices and storage rates; several Reserved Cluster and long-term discounted entries require contacting sales, so full long-term pricing details are not posted on the linked official pages.
More about capabilities and limits
Documented strengths
Wide global footprint and many GPU SKUs let you place workloads in regions to reduce latency.
Serverless endpoints claim scale-to-zero behavior with very low cold-start latency, useful for traffic that bursts from zero.
Public model endpoints let teams call pre-deployed models via API without provisioning GPUs.
Limitations to consider
Reserved cluster and long-term discounted pricing are not published on the public pages and require contacting sales.
Compliance coverage and guarantees can differ by region and deployment model, so you must verify region-specific compliance availability.
Choosing a plan
If you plan sustained capacity, confirm reserved-cluster discounts, per-second vs per-hour billing, and region-specific compliance guarantees with Runpod sales before committing to long-term capacity.
Based on the linked product documentation.
Available on
Web.
Official sources
Alternatives to Runpod
Choose around the work you need to do.
Read the full Runpod alternatives guide
Lambda
Use Lambda to access on-demand GPU instances and reserved clusters with documented per-hour pricing and single-tenant hardware for model training and high-density inference workloads.
