GPU compute for training and serving
Run workloads on NVIDIA architectures including Vera Rubin NVL72, Blackwell, and Hopper, with bare-metal and Kubernetes-native environments identified on the platform.
CoreWeave
Freemium
CoreWeave is an AI-focused cloud platform that combines GPU infrastructure, storage, networking, orchestration, and operational tooling for training and serving AI workloads. It supports teams moving from model experiments to production systems, including reinforcement-learning and agent-development workflows.
¿Qué es CoreWeave?
CoreWeave is an AI-focused cloud platform that combines infrastructure and development tooling for teams building and operating AI systems. Its platform covers GPU compute, AI-oriented storage, cluster networking, scheduling, and operational services, with support for both bare-metal and Kubernetes-native environments.
The product is organized around two connected layers. CoreWeave Infrastructure provides the foundation for training and serving workloads, while CoreWeave Forge supports the iterative AI loop: running systems, observing behavior, curating signals, improving a model or agent, and evaluating the next version. This makes the platform relevant to teams working across research, training, reinforcement learning, inference, and agent development.
CoreWeave also offers operational tooling for cluster health and workload visibility, plus CoreWeave ARENA for running real workloads on purpose-built infrastructure before production commitment. Pricing materials describe regional on-demand and spot GPU capacity, along with CPU and other cloud services.
¿Qué puede hacer CoreWeave?
AI-oriented storage
CoreWeave AI Object Storage and distributed file storage are designed for AI data throughput, with zero-egress migration described for storage workflows.
Cluster-scale networking
NVIDIA Quantum InfiniBand and Spectrum-X Ethernet provide networking options intended for large-scale cluster expansion and distributed training.
Managed workload orchestration
CoreWeave Kubernetes Service and SUNK, described as Slurm on Kubernetes, provide scheduling options for teams operating GPU workloads.
Infrastructure operations and observability
Mission Control automates node lifecycle and health checks and provides visibility into infrastructure and workload behavior.
AI iteration tools
Forge connects runs, production signals, curation, model or agent improvement, and evaluation; its ARIA agent analyzes runs and recommends the next iteration.
Casos de uso
“Large-scale model training”
Use GPU compute, high-throughput storage, cluster networking, and scheduling to run distributed or long-running training jobs with infrastructure health visibility.
“Reinforcement learning from production feedback”
Connect production signals with subsequent model improvement through Forge workflows designed around repeated feedback and training cycles.
“Model and agent iteration”
Track experiments, investigate which changes affected results, and use ARIA recommendations to plan the next model or agent revision.
“Production inference and serving”
Deploy serving workloads on available NVIDIA GPU configurations. The pricing page distinguishes inference usage for eligible customers, with availability and rates varying by configuration and region.
“Infrastructure evaluation”
Run real workloads through CoreWeave ARENA on purpose-built infrastructure before deciding whether to move to production.
Preguntas frecuentes
What does CoreWeave provide?
CoreWeave provides an AI cloud platform spanning GPU compute, AI-oriented storage, cluster networking, managed Kubernetes and Slurm-on-Kubernetes scheduling, and operational tooling. Forge adds tools for tracking and improving the AI development loop.
Can teams use CoreWeave for both training and inference?
Yes. The site describes infrastructure for training and serving AI workloads, and the pricing page lists on-demand and spot GPU instances as well as inference pricing for eligible customers. Specific availability and pricing depend on the hardware configuration and region.
Which hardware is available?
The platform identifies NVIDIA Vera Rubin NVL72, Blackwell, and Hopper architectures. The pricing page lists configurations including GB300 NVL72, GB200 NVL72, HGX B300, HGX B200, RTX PRO 6000 Blackwell Server Edition, GH200, H100, H200, L40, L40S, and A100 instances.
How can a team assess the infrastructure before production?
CoreWeave ARENA is presented as an evaluation environment where teams can run real workloads on purpose-built infrastructure before making a production commitment.
Datos rápidos
- Category
- AI cloud infrastructure
- Primary users
- AI research, engineering, and platform teams
- Core workflow
- Run workloads, observe results, curate signals, improve models or agents, and evaluate new versions
- Compute
- NVIDIA GPU instances across Vera Rubin, Blackwell, and Hopper families
- Deployment options
- Bare metal and Kubernetes-native environments
- Pricing model
- Regional on-demand and spot GPU pricing, with additional CPU and other service pricing listed separately
Alternativas a CoreWeave
UbiOps
ubiops.com
Model serving and orchestration for AI teams deploying workloads across cloud and private infrastructure.
Ray
www.ray.io
Ray is an open-source, Python-native framework for distributing and scaling AI, machine learning, and other compute-intensive workloads. It helps developers run workloads from a laptop to clusters with thousands of GPUs and supports data processing, training, serving, inference, and reinforcement learning.
Lambda
lambda.ai
Lambda provides cloud GPU compute for AI training, fine-tuning, inference, and prototyping. Teams can launch on-demand GPU instances, use production-ready 1-Click Clusters, or discuss reserved and single-tenant infrastructure for larger workloads.
FriendliAI
friendli.ai
FriendliAI is an inference cloud for deploying frontier open-weight and custom AI models in production. It offers serverless Model APIs, dedicated GPU endpoints, and BYOG options for agent, multimodal, and other high-throughput workloads.
dstack
dstack.ai
A control plane for provisioning and orchestrating GPU workloads across cloud, Kubernetes, and on-prem infrastructure.
DeepInfra
deepinfra.com
DeepInfra provides hosted machine-learning model inference and on-demand GPU instances for developers and teams. Its catalog covers text, image, audio, video, embedding, reranking, and other model workloads with pay-as-you-go pricing.