CoreWeave
Large-scale GPU infrastructure provider running Kubernetes-native compute, aimed at organisations training and serving models at production scale.
Pricing Plans
| Plan | Price/mo | CPU | RAM | Notes |
|---|---|---|---|---|
| A40 48GB | — | 16 vCPU | 128GB | Around $1.28/hr |
| A100 80GB NVLINK | — | 24 vCPU | 256GB | Around $2.39/hr |
| HGX H100 node | — | 128 vCPU | 2TB | Contract pricing, 8 GPUs per node |
· Updated
About CoreWeave
CoreWeave began life mining cryptocurrency and pivoted the GPU fleet into an AI cloud, which turned out to be extraordinarily well timed. It is now one of the largest dedicated GPU providers outside the hyperscalers, and its customers are mostly companies training or serving large models rather than individuals.
The architecture is Kubernetes-native. You are not really renting VMs; you are scheduling workloads onto a managed cluster. If your team already deploys with Kubernetes manifests, this feels natural. If you wanted to SSH into a box and run a script, it does not.
Pricing
Per-GPU hourly rates are published and competitive: an A40 sits around $1.28/hr and an A100 80GB with NVLINK around $2.39/hr. CPU and RAM are billed separately and granularly, so the true cost of an instance is the sum of its parts rather than a single sticker number.
Serious H100 capacity is contract based. This is not a provider you sign up to on a whim with a credit card and one GPU, and the onboarding reflects that.
Service Details
Storage is network-attached, backed by a distributed filesystem, and there is object storage with an S3-compatible API. Networking between GPU nodes uses InfiniBand on the HGX tiers, which matters once a training job spans more than one machine.
The images are Linux, orchestration is Kubernetes, and the tooling assumes you know both. Our k3s and Kubernetes guide and the container basics guide are reasonable starting points if that stack is new.
Best fit: a funded team with a real training or inference workload and someone on staff who is comfortable with cluster operations. Poor fit: one person who wants to fine-tune a model this weekend, where RunPod or Vast.ai will be far less friction.
Available in: United States, United Kingdom, Norway, Sweden, Spain
Learn More
For more information about CoreWeave, visit their website at https://www.coreweave.com/