Compute
CoreWeave offers compute options for running containerized AI and HPC workloads:CoreWeave Kubernetes Service (CKS)
Managed Kubernetes on bare metal for training, inference, and HPC.
SUNK
Slurm on Kubernetes for batch and burst workloads.
Inference
Serverless and dedicated inference options for serving AI models.
CoreWeave Sandbox
Ephemeral compute sandboxes for agents and code interpretation.
Storage
CoreWeave storage solutions support the data requirements of AI and ML workloads:CoreWeave AI Object Storage
Purpose-built, S3-compatible object storage for training datasets, model weights, checkpoints, and other distributed bulk data.
Distributed File Storage
POSIX-compliant shared filesystem with high throughput and low latency for multi-Node access.
Dedicated VAST Storage
Single-tenant VAST clusters co-located with your GPU infrastructure for petabyte-scale capacity and full cluster control.
Local Storage
High-performance, Node-local NVMe storage for temporary scratch space, caching, and logs.
CoreWeave Database
Managed, PostgreSQL-compatible database service that runs inside your CKS cluster.
Networking
CoreWeave networking products create secure, high-performance connections between your resources and services:Virtual Private Clouds
Isolated networks that connect your CoreWeave Kubernetes Service (CKS) clusters and resources while keeping traffic private and secure.
HPC Interconnect
High-performance, RDMA-enabled fabrics for GPU-to-GPU communication over InfiniBand or Spectrum-X (RoCE).
Direct Connect
Private, dedicated links to CoreWeave through Equinix or Megaport, without adding physical ports or a network presence in our regions.
IP Addresses
Public IPv4 addresses and Bring Your Own IP (BYOIP) for exposing your Services to the internet.
Ingress Service
Exposes Services running on CKS to the public internet using load balancers and public IPv4 addresses.
Nimbus
Control and data-plane software built into each Node’s BlueField DPU for networking and tenant isolation.