Google Cloud Managed Lustre Dynamic Tier costs $0.06/GB-month, scales to 80 PB, and provides sub-millisecond latency for hot data.

Google Cloud prices Managed Lustre Dynamic Tier at US$0.06 per GB-month, with a single namespace for hot and colder data. Google says throughput scales linearly with capacity up to 80 petabytes (PB) and hot-data latency remains below one millisecond as the service scales to tens of thousands of clients.

The latency figures differ by storage tier. The High-Performance Cache, using solid-state drives (SSDs), provides sub-millisecond reads, while the Capacity Pool, labeled “HDD” for hard-disk storage and built on Google Cloud Hyperdisk throughput, has an average read latency of 10–30 milliseconds.

Google describes the price as a single flat fee, with no independent charges for disk-media types, data movement within the namespace, or metadata input/output operations per second (IOPS). Its stated objective is to reduce separate storage environments, manual data staging, dataset copying, and disjointed namespaces across AI and high-performance computing (HPC) workflows.

Recommended workloads include multi-epoch training, optimized-size data fetching, write-heavy checkpointing, and rapid checkpoint restoration: hot training data is promoted after its first run, while checkpoint writes land in the cache and older checkpoints are demoted to the capacity pool. Google also cites approximately 300-microsecond average read latency for interactive development tasks such as repository cloning, library compilation, and notebooks.