CoreWeave
AI cloud with on demand NVIDIA GPUs, fast storage and orchestration, offering transparent per hour rates for latest accelerators and fleet scale for training and inference.
Overview
CoreWeave provides specialized GPU infrastructure across regions with options from A100 to B200 and GB200 class systems. The public pricing page lists hourly rates for select SKUs with on demand access, while larger reservations and multi node clusters are handled by sales. Users deploy via Kubernetes, scale jobs elastically and attach high performance storage.
The platform targets model training, fine tuning and real time inference, and is used by enterprises and labs that need capacity without building data centers. Billing is usage based with per hour GPU prices, plus storage and networking. Documentation covers templates and best practices for throughput and reliability.
Key features
- On demand NVIDIA fleets including B200 and GB200 classes
- Per hour pricing published for select SKUs
- Elastic Kubernetes orchestration and job scaling
- High performance block and object storage
- Multi region capacity for training and inference
- Templates for LLM fine tuning and serving
- Private networking and security options
- Support for reservations and larger clusters
Best for
- Spin up multi GPU training clusters quickly
- Serve low latency inference on modern GPUs
- Run fine tuning and evaluation workflows
- Burst capacity during peak experiments
- Disaster recovery or secondary region runs
- Benchmark new architectures on latest silicon
- Cost model comparisons across GPU SKUs
- Hybrid setups with on prem plus cloud overflow
Capabilities
On Demand GPUs
Launch B200 or GB200 class instances on demand with per hour pricing and scale capacity up or down for experiments or production.
Kubernetes & Storage
Run jobs on managed Kubernetes with high performance storage so data pipelines keep GPUs saturated.
Right Sizing & Regions
Choose GPU, memory and region combinations that meet SLA and cost targets for each workload.
Reservations & Support
Work with sales for long term reservations, multi node clusters and support plans that match roadmap and budgets.
Frequently Asked Questions
What is the starting price per hour?
Published examples show GB200 NVL72 from forty two dollars per hour and B200 from sixty eight dollars eighty per hour, other SKUs vary by region and configuration.
Can I reserve capacity for months?
Yes, contact sales for committed reservations, dedicated clusters and custom networking.
How do I run training jobs?
Use templates and Kubernetes, attach storage and scale workers, then monitor throughput to keep GPUs busy.
Is there a free tier?
There is no free tier, billing is usage based for compute, storage and egress.
Do you support multi region redundancy?
Yes, customers deploy in multiple regions for resilience and traffic distribution.
Can I bring my own images and frameworks?
You can run common ML stacks or custom containers with your dependencies.
How is security handled?
Private networking, isolation and controls are provided, along with enterprise agreements for regulated workloads.
Where do I see all GPU prices?
A live pricing page lists select SKUs, contact sales for configurations not shown.
