CoreWeave GPU Pricing, Specs & Review
Last updated: 2026-03-17
Overview
CoreWeave is a specialized GPU cloud provider purpose-built for AI and HPC workloads. Built on Kubernetes-native infrastructure, CoreWeave offers bare-metal GPU performance with cloud flexibility. They provide NVIDIA B200, H200, H100, A100, and L40S GPUs with InfiniBand interconnects for distributed training. CoreWeave is known for competitive pricing compared to hyperscalers, with a focus on GPU-intensive workloads rather than general-purpose cloud computing.
GPU Pricing
| GPU | VRAM | On-Demand $/hr | Spot $/hr | Monthly |
|---|---|---|---|---|
| B200 (8x) | 8x 192 GB | $68.80 | N/A | ~$49,536 |
| H100 HGX (8x) | 8x 80 GB | $49.24 | N/A | ~$35,453 |
| H100 PCIe | 80 GB | $4.76 | N/A | ~$3,427 |
| A100 80GB NVLink | 80 GB | $2.21 | N/A | ~$1,591 |
| A100 80GB (8x) | 8x 80 GB | $21.60 | N/A | ~$15,552 |
| L40S (8x) | 8x 48 GB | $18.00 | N/A | ~$12,960 |
Key Features
- Kubernetes-native GPU cloud infrastructure
- NVIDIA B200, H200, H100, A100, and L40S GPUs
- InfiniBand networking for multi-node training
- Bare-metal GPU performance with cloud flexibility
- Per-minute billing with no minimum commitment
- Custom reserved pricing for long-term workloads
- Managed Kubernetes with GPU scheduling
- Purpose-built for AI/ML and HPC workloads
Pros & Cons
Pros
- Competitive pricing vs. hyperscalers (30-50% lower)
- Purpose-built GPU infrastructure with high availability
- Native Kubernetes support with GPU-aware scheduling
- InfiniBand networking for distributed training
- Strong GPU availability and rapid provisioning
Cons
- Fewer regions than hyperscale cloud providers
- No spot/preemptible instance pricing
- Less mature ecosystem compared to AWS/Azure/GCP
- Limited general-purpose cloud services
FAQ
What GPU models does CoreWeave offer?
CoreWeave offers NVIDIA B200, H200, H100 SXM, H100 PCIe, A100 80GB NVLink, L40S, and RTX A6000 GPU instances optimized for AI training and inference.
What is the cheapest GPU option on CoreWeave?
CoreWeave A100 80GB NVLink instances start at $2.21/hr per GPU. Full 8-GPU H100 HGX nodes are available at approximately $6.15/hr per GPU.
Does CoreWeave offer reserved GPU pricing?
Yes. CoreWeave offers reserved pricing with discounts for longer-term commitments. Contact their sales team for custom reserved instance pricing.