CoreWeave
6.7/10CoreWeave is a GPU-focused cloud that rents NVIDIA GPU and CPU instances, managed Kubernetes, and storage to teams training and serving AI models.
Data last updated November 2026 · Our methodology
Overview
CoreWeave is a specialized AI cloud. It rents NVIDIA GPU capacity, including GB200 NVL72, HGX B200, H200, H100, A100, and L40S, as well as AMD and Intel CPU instances, mostly on bare-metal servers. Workloads run through CoreWeave Kubernetes Service (CKS), whose control plane is free, or through SUNK, its Slurm-on-Kubernetes scheduler. The platform also offers AI Object Storage with hot, warm, cold, and archive tiers, distributed file storage, InfiniBand networking for multi-node training, managed inference, and observability tooling. CoreWeave owns Weights & Biases and packages training and inference tooling as CoreWeave Forge. Pricing is metered by the hour per instance. For example, an 8-GPU H100 node is $49.24 per hour on demand and $19.71 per hour on spot. Committed contracts can cut on-demand rates by up to 60%. Data transfer, including internet egress, is free, while public IPs cost $4 per month. Infrastructure can be managed as code with a Terraform provider that also works with OpenTofu and covers clusters, VPCs, storage buckets, and inference resources. The Trust Center lists SOC 2 Type II, ISO 27001, ISO 27017, and ISO 27018. The best fit is AI labs, enterprises, and well-funded startups running large training or inference jobs. The main limitation is breadth: CoreWeave has no free tier, no managed databases, no serverless functions, and no CDN, and its regions are concentrated in North America and Europe, so most teams still need a second cloud for general application hosting.
Our Rating
Hosting & Infrastructure6.9/10
AI Software6.4/10
Who CoreWeave is best for
Good fit
Built for large AI training clusters, reserved multi-year contracts, and dedicated Direct Connect links.
ML engineers can use kubectl, Helm, Slurm, and Terraform, though the platform offers little for general web app development.
Less suited
AI startups get on-demand GPUs without a minimum commitment, but there is no free tier and the best rates require contracts.
There is no conventional hosting, and minimum spend is reported to be high for light use.
Pros & Cons
Pros
- Free internet egress and internal data transfer, which lowers the cost of moving large datasets and model checkpoints.
- Large inventory of current NVIDIA hardware, from L40S to GB200 NVL72, with InfiniBand for multi-node training.
- Spot rates are well below on-demand rates, for example $19.71 versus $49.24 per hour for an 8-GPU H100 node, and committed discounts reach up to 60%.
- The CKS control plane and SUNK Slurm scheduler are free, so you pay only for compute.
- A Terraform and OpenTofu provider manages clusters, VPCs, storage, and inference as code.
Cons
- No free tier or trial credits, and minimum spend is reported to be high even for light use.
- No managed databases, serverless functions, or CDN, so a second cloud is often needed.
- Regions are concentrated in North America and Europe, with limited Asia-Pacific coverage.
- The best pricing usually requires multi-year contracts and an enterprise sales process.
- Few public user reviews are available on G2 or Gartner to validate claims.
Key Features
Not included
- Content Delivery Network
- Managed Databases
- Managed WordPress
- Serverless Functions
- Shared Hosting
CoreWeave Pricing
On-Demand
- Hourly billing per instance
- GB200 NVL72 at $42.00 per hour
- Free CKS control plane and SUNK
- Free egress and internal data transfer
- HGX B200 8-GPU node at $68.80 per hour
- HGX H100 8-GPU node at $49.24 per hour
Reserved Capacity
- Quote from sales
- Committed-use contracts
- Up to 60% off on-demand rates
Spot
- Interruptible capacity
- A100 8-GPU node at $9.65 per hour
- HGX H100 8-GPU node at $19.71 per hour
Frequently Asked Questions
No. CoreWeave lists data transfer to the internet and within CoreWeave as free. Public IPs cost $4 per month each.
Independent reviews report that CoreWeave has no free tier or trial credits.
CoreWeave says committed usage can be up to 60% cheaper than on-demand rates. Reserved pricing is quoted by sales.
Its Trust Center lists SOC 2 Type II, ISO 27001, ISO 27017, and ISO 27018. Reports are available on request.
A third-party directory reports a 99.9% monthly compute objective for multi-region deployments, 99% for a single region, and service credits for outages.
Yes. The official provider manages CKS clusters, VPCs, AI Object Storage buckets and policies, and Managed Inference resources, and it also works with OpenTofu.
Similar Tools
See all alternatives
Amazon's public cloud platform, with over 200 pay-as-you-go services for compute, storage, databases, networking and AI, used by startups through large enterprises and governments.
Google's public cloud platform providing pay-as-you-go virtual machines, Kubernetes, serverless, databases, storage, networking, data analytics and AI services for businesses of every size.
Northflank is a Kubernetes-based developer platform for deploying services, jobs, databases and GPU workloads on its managed cloud or on customers' own cloud accounts.