From $0.02 usage-based🇺🇸
Inference API for more than 100 open and commercial models billed per token, plus GPU instances and dedicated clusters.
CoreWeave AI features and pricing (October 2026)
GPU cloud for training and serving large AI models, with a managed Kubernetes service, Slurm integration and managed inference.
Usage-based
from $6.50 usage-based
Pricing details
Hourly on-demand and spot prices per instance; according to the pricing page from $6.50 per hour for one NVIDIA GH200, while eight NVIDIA L40 GPUs cost $10.00 per hour ($1.25 per GPU), as of September 2026.
What is CoreWeave?
CoreWeave is a GPU cloud built for AI workloads. It rents NVIDIA GPU capacity for model training and inference to AI labs, software companies and enterprises that need large clusters.
CoreWeave offers GPU and CPU compute, bare metal servers, storage and networking, plus a managed Kubernetes service whose control plane the pricing page lists as free. Platform tools such as Tensorizer for faster model loading, observability and SUNK, which combines Slurm and Kubernetes for training, sit on top. The company is based in Livingston, New Jersey, lists data center regions in North America and Europe on its pricing page and publishes a trust center that also covers Weights & Biases, which belongs to CoreWeave.
CoreWeave AI features and use cases
- Serves open-source models per tokenServerless Inference runs models from a curated open-source catalog and bills per token, with built-in tracing. Teams can test a model in an application without provisioning GPUs themselves.
- Hosts custom and fine-tuned modelsDedicated Inference deploys your own model weights on the GPUs you choose, with routing, scaling and lifecycle managed by CoreWeave. This suits teams that need predictable latency for one model.
- Runs large training jobsSUNK lets research teams submit Slurm jobs on Kubernetes-managed GPU clusters. Platform teams keep one cluster for training and serving instead of maintaining two separate schedulers.
What sets CoreWeave apart
- Three inference tiers: serverless per token, dedicated endpoints and self-managed inference on its Kubernetes service.
- Trust center lists ISO/IEC 27001, 27017, 27018 and SOC 2 Type II, with a public subprocessor list.
Where CoreWeave is strongest
United States Available internationally
CoreWeave is headquartered in Livingston, New Jersey, in the United States.
Questions to ask before you buy CoreWeave
Clarify which region your data and model weights will be stored in, since the pricing page names North America and Europe without a finer breakdown. Hourly GPU prices add up quickly for idle instances, so plan reservations or spot use. Request the security documentation through the trust center and check the data processing terms before sending personal data to inference endpoints.
CoreWeave reviews
No reviews yet. Be the first to share how your team uses it.
Every review shows its verification level. We read each one before it goes live. How we check reviews
Questions about CoreWeave
How does CoreWeave charge for GPUs?
By the hour per instance, with on-demand and spot prices on the pricing page. In September 2026 the lowest instance price was $6.50 per hour for one NVIDIA GH200; eight NVIDIA L40 GPUs cost $10.00 per hour, or $1.25 per GPU.
Does CoreWeave offer serverless inference?
Yes. Serverless Inference bills per token for models from a curated open-source catalog, and Dedicated Inference runs custom models on selected GPUs.
Which certifications does CoreWeave hold?
The trust center lists ISO/IEC 27001, ISO/IEC 27017, ISO/IEC 27018 and SOC 2 Type II. Detailed reports are available on request through the trust center.
CoreWeave update historyno updates yet
No changes recorded for CoreWeave yet.
Changes the vendor made and confirmed user reports, each checked against the vendor's own pages. We review every tool monthly.
Alternatives to CoreWeave
Tools in the same category (GPU rental) with the most overlapping markets.
Usage-based🇺🇸
GPU cloud with low-cost hourly Pods, serverless endpoints that scale to zero and clusters for developers and small teams.
From $1.50 usage-based🇺🇸
US AI cloud on its own data centers, with NVIDIA and AMD GPUs, managed Kubernetes and Slurm, and managed inference.
From $0.79 usage-based🇺🇸
GPU cloud offering single instances, 1-Click Clusters and Superclusters on NVIDIA hardware for AI training and inference.