Google Cloud announced a new storage solution: Rapid Storage. This so-called zonal bucket within Cloud Storage offers fast and consistent access to data, with read and write latencies below one millisecond.
According to Google, this storage class is up to 20 times faster and supports throughput speeds of up to 6 TB per second. This is reported by InfoQ. Unlike traditional Cloud Storage buckets, which distribute data across multiple regional zones, Rapid Storage stores all data within a single zone. This allows applications that use GPUs and TPUs to benefit from higher throughput and lower latency.
According to Google managers Sameet Agarwal and Asad Khan, this approach aims to train and control AI models more efficiently. They emphasize that traditional object storage is often too slow for these applications. The Colossus file system addresses this problem by bringing storage closer to computing power.
The new storage can be mounted as a file system via Cloud Storage FUSE. This provides support for popular AI frameworks such as TensorFlow and PyTorch. In a technical article, Google engineers Denis Serenyi and Vivek Saraswat explain that Rapid Storage uses a gRPC streaming protocol and features such as object replenishment. This is achieved without compromising scale or performance.
Different forms of object storage
On Reddit, Rapid Storage was compared to Amazon S3 Express. It was noted that Google is now the only major cloud provider to offer different forms of object storage, ranging from fast single-zone to double-replicated regional storage, all with a single, uniform API.
At the same time, AWS announced that it is lowering prices for S3 Express One Zone. Storage costs will drop by 31%, costs for PUT requests by 55% and for GET requests by 85%. According to Agarwal and Khan, Rapid Storage still offers significant advantages. These include five times lower latency for random read and write operations compared to other major cloud providers. In addition, it offers throughput of up to 6 TB/s per bucket and support for up to 20 million requests per second.
Rapid Storage wasn’t the only new storage product focused on AI at Cloud Next 2025. Google also introduced Anywhere Cache, an SSD cache that accelerates regional buckets by storing data locally, and Google Cloud Managed Lustre, a powerful file system based on DDN EXAScaler Lustre. This system is designed for AI workloads and offers petabyte-scale performance with millions of IOPS and sub-millisecond latency.
Rapid Storage is currently available as a preview.