CoreWeave AI Object Storage
CoreWeave AI Object Storage is purpose-built to serve large training datasets, model weights, checkpoints, and other distributed bulk data through an S3-compatible API. Object Storage is platform independent, enabling applications to connect across cloud providers. It provides authentication, security policies, read-after-write consistency, versioning, bucket lifecycles, and other features that are essential for managing the large datasets required for AI and ML workloads. Object Storage serves data directly to CPU and GPU Nodes, or through model serializers such as CoreWeave Tensorizer. Performance is boosted further with LOTA (Local Object Transport Accelerator), an intelligent proxy for Object Storage. LOTA is installed on every GPU and CPU Node to provide an efficient, local gateway to Object Storage that caches fetched data to reduce load times. By keeping recently accessed data on the local Node, LOTA reduces the time it takes to load data from storage, making it well-suited to AI and ML workloads that require fast access to large datasets.Distributed File Storage
Distributed File Storage is a POSIX-compliant shared filesystem that allows multiple GPU Nodes to access the same data simultaneously. Distributed File Storage has high throughput and low latency, provides automatic data snapshots, and has asynchronous file deletion features. Distributed File Storage is well-suited to applications that require synchronization between Pods, such as distributed training, because it minimizes performance bottlenecks caused by file contention.Dedicated VAST Storage
Dedicated VAST Storage provides a single-tenant VAST cluster co-located with your GPU infrastructure in a CoreWeave data center. Each cluster is physically isolated to a single tenant, with direct access to the VAST Management System (VMS), multi-protocol support (NFS, S3, NVMe/TCP, and SQL), and advanced data services including VAST Catalog, DataBase, DataEngine, and cross-cluster replication. Dedicated VAST is designed for workloads that need petabyte-scale capacity, full cluster management control, and features such as custom QoS policies, configurable snapshots, and per-tenant encryption keys.Local Storage
Local Storage is available on all GPU Nodes for high-performance temporary storage, caching, and logs. This non-persistent local storage isn’t intended for long-term data retention. The amount of Local Storage available on Nodes varies depending on the Node’s instance type.Choose the right storage
Most workloads use more than one storage type. Choose storage based on data class and I/O profile: keep the active temporary working set on Local Storage, durable portable artifacts in Object Storage, and POSIX-active data in DFS. Does your data need to outlive the Pod or Node? If not, use Local Storage. Local Storage provides NVMe-backed scratch space for caches, logs, and intermediate artifacts, but it does not survive Pod deletion or Node reboots. If yes, continue. What interface and I/O pattern does your application use? Use CoreWeave AI Object Storage when your application uses an S3-compatible API and primarily reads or writes large objects with parallel, contiguous access, such as training datasets, model weights, checkpoints, backups, or data that must be accessible across clusters or availability zones. Use LOTA from within CoreWeave clusters to cache hot objects on local Node NVMe. Use Distributed File Storage when your application requires a POSIX filesystem, such as persistent home directories, shared workspaces, frequent file updates or directory listing, or small and random I/O.DFS cannot be mounted from another cluster, availability zone, or region. If an application requires both POSIX access and multi-cluster data sharing, use Object Storage as the durable source of truth and stage data to DFS or Local Storage within each cluster.
Need petabyte-scale capacity or full cluster control?
Dedicated VAST Storage provides a single-tenant VAST cluster co-located with your GPU infrastructure, with multi-protocol access (NFS, S3, NVMe/TCP, and SQL) and petabyte-scale capacity. Dedicated VAST is not self-serve: provisioning requires coordination with CoreWeave. Contact Sales to discuss sizing and availability.