Managed Lustre

Distributed file system for ML.

Fully managed Lustre file system for AI/ML workloads. Scales to exabytes with automatic tiering to object storage. Zero operational overhead.

Managed Lustre1 TB/s ThroughputS3 Data RepositoryAuto-import · Lazy loadingGPU 1GPU 2GPU 3GPU 41 TB/s

1 TB/s

Speed

Exabytes

Scale

Fully

Managed

S3 auto

Tiering

TB/s for ML training.

Managed Lustre with S3 data import.

Terabyte/s throughput

Sustained 1+ TB/s for training data loading. Scales linearly with capacity.

GPU-optimized

Designed for ML training workflows. Feed thousands of GPUs with training data.

Auto-tiering

Cold data automatically tiers to object storage. Hot data stays on NVMe.

Zero ops

No servers to manage. Automatic scaling, patching, and failover.

S3 data import

Import existing training data from S3 buckets. Lazy loading for instant availability.

Snapshot and clone

Instant snapshots for experiment reproducibility. Clone datasets in seconds.

Getting started

Launch your first instance in three steps. CLI, console, or API — your choice.

Terminal
ur storage lustre create ml-data \
  --capacity=50TB \
  --import-path=s3://training-data

High-throughput workloads.

ML training and genomic analysis.

Model training storage

Feed GPU clusters with TB/s of training data.

View tutorial

Suggested configuration

1 TB/s · S3 import · Snapshot

Estimate your costs

Create detailed configurations to see exactly how much your architecture will cost. Pay for what you use, down to the second.

Configuration 1

Estimated: $8.73/mo

Lustre File System

Capacity

TB

Operations & Transfer

GB
Config 1 cost$8.73

Cost details

$8.73

Zero ops Lustre. Auto-tiering to object storage.

Configuration 1
$8.73
10 TB Storage$0.23
Data Egress$8.00
Operations$0.50

Works seamlessly with

GPU Instances
Object Storage
HPC Clusters
Monitoring
IAM
Logging

Frequently asked questions

Lustre, fully managed.

TB/s throughput for ML training. Zero operational overhead.