Storage

Get performant, secure, and reliable storage for AI.

Tailor-made for AI workloads

Don’t let storage performance slow your cluster down.

Get higher performance for containerized workloads and virtual servers.

Feed data into GPUs ASAP

GenAI models need a lot of data—and they need it fast. Handle massive datasets with reliability and ease, helping to enable better performance and faster training times.

Pick up where you left off

Keep your teams on track. Reduce major delays after hardware interruptions with storage that keeps GPUs working while checkpoints commit.

Get top-tier reliability

Keep production on track and avoid massive data losses. Automated snapshots occur every 6 hour interval with a 3 day retention, meaning your teams can rest assured their work is saved. Plus, we manage storage separately from compute, making data easier to move and track.

Stay secure

CoreWeave Storage follows industry best practices with security. Encryption at rest and in transit, identity access management protocol, authentication, and policies with roles-based enable data protection and security.

CoreWeave AI Object Storage

Train, fine-tune, and deploy models faster and more reliably at any scale. Eliminate replication costs and performance tradeoffs with AI-native object storage.

See how it works

Local Object Transport Accelerator (LOTA)

LOTA brings data closer to your GPUs by caching directly on GPU nodes, delivering near-local throughput of up to 7 GB/s per GPU. The result: faster training, lower latency, and efficiently utilized clusters without complex replication layers.

Automated, usage-based billing

Seamlessly tier inactive data at a lower rate while keeping it accessible when needed, lowering TCO without impacting performance. No more manual tiering or multiple APIs. Just simple pricing without egress, ingress, or request fees.

Unified data across regions and clouds

Simplify operations, cut costs, and keep GPUs efficiently utilized with a single global dataset and no replication. LOTA accelerates reads, and cross-region write acceleration brings writes to local latency across regions.

Industry-leading performance

Keep compute efficiently utilized with throughput up to 7 GB/s per GPU scaling linearly with cluster size, far beyond traditional object storage.

Simplified scaling

One global dataset and comprehensive observability tools make scaling effortless wherever your workloads are.

‍

Lower TCO

Reduce costs with automated usage-based billing for inactive data and transparent, straightforward pricing.

Enterprise-grade reliability

Protect critical AI workloads with SSO/SAML authentication, 11 nines durability, 99.9% uptime, and encryption at rest and in transit.

During training, keeping GPUs from data starvation is a delicate balance of minimizing storage latency without complicating storage operations. Caching on the compute node itself, across thousands of nodes in a cluster, is a meaningful architectural difference from zone-level approaches. Doing it without application changes or a replication pipeline to maintain is what makes it operationally viable at scale.

Dave McCarthy

Group Vice President, IDC

Zero Egress Migration ([0]EM)

The CoreWeave [0]EM program significantly reduces the cost and complexity of large-scale transfers, letting you move data directly from any major cloud into CoreWeave AI Object Storage with no egress fees*, no exit penalties, and minimal disruption to your innovation.

Learn more

CoreWeave [0]EM program

$0.00/GB

in egress fees

Transfers from major cloud providers
Fully-managed migrations
End-to-end validation
No required account closures
Direct-to-expert support
No egress fees
No vendor lock-in
No exit penalties

Legacy Hyperscalers

Up to

$0.12/GB

in egress fees

Transfers from major cloud providers
Fully-managed migrations
End-to-end validation

No required account closures
Direct-to-expert support
No egress fees
No vendor lock-in
No exit penalties

*With a new minimum two-year CoreWeave storage contract, you can do a one-time migration at no charge of up to 4.17 GB per USD of total contract value, via a third-party provider selected by CoreWeave.

Distributed file storage

CoreWeave distributed file storage helps with centralized asset storage or parallel computation setups necessary for GenAI.

See the benchmarks Optimize your storage

A network made for AI

Our highly performant, ultra-low latency networking architecture is built from the ground up to get data to GPUs with the speed and efficiency AI models need to develop, train, and deploy.

Strategic partnerships

Our partnership with VAST enables us to manage and secure hundreds of petabytes of data at a time. Plus, we’re POSIX-compliant and suitable for shared access across multiple instances.

Ultra-fast access at scale

Leverage a petabyte-scale shared file system with up to 1 GB/s per GPU. Our benchmarks show strong results for up to 64-node NVIDIA H200 GPU clusters.

Local storage

Our Kubernetes solution uniquely allows customers to access local storage. CoreWeave supports ephemeral storage up to 60TB, depending on node type.

All physical nodes have SSD or NVMe ephemeral (local) storage. No need for Volume Claims to allocate ephemeral storage. Simply write anywhere in the container file system.

That’s all included at no additional cost.

Dedicated storage clusters

Get the specialized storage you need from the providers you trust.

CoreWeave delivers dedicated clusters from best-in-class AI storage partners—including VAST Data, WEKA, DDN, IBM Storage Scale, and EverPure—to give your workloads the performance, isolation, and security they demand, all on the top-ranked AI cloud.

Copied

On-Demand Webinar

How to Move Beyond Tiers, Tradeoffs, and Runaway Costs in AI Storage

Stop overpaying for AI storage or struggling with tiers and complexity. Discover how CoreWeave keeps every dataset instantly available and performant while cutting costs up to 65%.

Watch now

Specialized storage for AI

CoreWeave infrastructure all works in tandem to ensure the future of GenAI.

Find out how