-
Rapid Cache, formerly Anywhere Cache, accelerates bandwidth for bursty workloads like model loading for inference, delivering an aggregate read throughput of 2.5 TB/s for existing buckets, with no code changes. The new ingest-on-write feature provides up to 2.2x faster checkpoint restores, allowing training clusters to recover faster from interruptions.
Rapid Cache’s combination of simplicity and performance has resulted in strong adoption, including from cutting-edge AI/ML customers like Thinking Machines Lab.
“Rapid Cache has become a core foundation of our AI/ML data infrastructure, supporting our critical workflows, from data prep and pretraining to training and model loading. By acting as a crucial bandwidth shield and booster, it enables us to scale our data-intensive workloads across our entire fleet without compromise, providing us with the on-demand high bandwidth and consistent stability that we need to innovate at speed.” – James Sun, Member of Technical Staff, Thinking Machines Lab
Google Cloud Managed Lustre
The Lustre parallel file system is the industry standard for organizations whose AI training and inference workloads require high throughput and sub-millisecond latency, and is trusted by AI labs and HPC centers worldwide to feed thousands of accelerators simultaneously and keep them saturated under pressure. Google Cloud Managed Lustre brings that capability as a fully managed service, and with today’s announcements, it is the most performant managed Lustre offering available in any cloud.
Managed Lustre now delivers up to 10 TB/s of throughput — a 10x increase since last year and 4–20x higher than managed Lustre offerings from other hyperscalers for a single instance. Powered by C4NX VMs and Hyperdisk Exapools, Managed Lustre writes and restores checkpoints 2.6x faster when compared to other Google Cloud storage solutions.
The new Dynamic tier ($0.06/GB-month) delivers the low-latency performance required for intense AI workloads like training and checkpointing. By serving data from persistent disk rather than relying on object-based caching, we eliminate a performance cliff — helping ensure your data remains responsive and your accelerators stay productive. A single SKU provides simple, predictable billing without the hidden complexity of traditional data tiering.
“By integrating Managed Lustre we eliminated the typical onboarding bottlenecks, allowing us to hit the ground running with the inferencing workload. This high-throughput, low-latency storage keeps our B200 GPUs fully saturated, driving a substantial performance gain in LLM inference over the H200. For our customers, this translates directly into faster, more responsive AI agents that can handle complex reasoning at a fraction of the previous latency.” – Lavnaya Karanam, Software Engineering PMTS, Salesforce
Smart Storage: Context for the AI era
The beauty of an object storage system like Cloud Storage has long been its simplicity: the system knows the object’s name, its size, and when it was created. But if you want to understand the object’s content — what entities it references, whether it contains sensitive PII, or whether it’s relevant to a pending query — you need to use custom pipelines, separate databases, and bespoke enrichment systems.
AI has changed the equation. To fine-tune a model, you need to select the right objects from the get-go, from a corpus of millions. Building an agent requires retrieving the right context for each decision. To meet a compliance obligation, you need to know what every file contains up front, before it becomes a liability. In each case, the bottleneck isn’t compute or model quality — it’s the inability to describe, find, and act on objects at scale.
To bridge that gap between stored and usable data, last year we introduced Smart Storage, a set of capabilities built directly into Cloud Storage that makes every object self-describing. New Smart Storage capabilities include:
-
Automated annotations, which eliminates the need to build and maintain custom annotation pipelines. With Smart Storage enabled, Cloud Storage can now automatically generate context — including image annotations — so your data is discoverable and usable from the moment it lands. You pay to annotate the data once at write time, and every downstream system can use those annotations immediately for the life of the object.
-
Cloud Storage MCP server lets you read, write, and analyze Cloud Storage data using the standard MCP protocol.
Smart Storage enables these capabilities, and others, thanks to its object context, now generally available. This metadata substrate adds structured, mutable, IAM-governed context to every object. Customers write their own tags and classifications; Google’s annotation pipelines automatically attach labels, extracted entities, and compliance signals.
