Google Cloud adds Cloud Storage Rapid with zonal Rapid Bucket and Rapid: what developers gain

Cloud Storage Rapid, unveiled at Google Cloud Next ’26, is a suite of object‑storage capabilities built for data‑intensive AI and analytics.

Google Cloud announced Cloud Storage Rapid at Google Cloud Next ’26, a new family of object‑storage capabilities designed to remove storage I/O bottlenecks for AI and analytics workloads by colocating high‑performance zonal storage with compute. Rapid Bucket is now generally available (formerly called Rapid Storage), and Rapid Cache joins the family to accelerate reads from existing buckets. This update targets teams running large‑scale training and inference on GPUs and TPUs, where storage stalls can leave accelerators idle.

Rapid Bucket is presented as a dedicated zonal object storage option that leverages Google’s Colossus backend to deliver block‑like latency while retaining object semantics. Google cites sub‑millisecond latency, up to 20 million queries per second, and more than 15 TB/s of aggregate read throughput from a single Rapid zonal bucket. The tier also adds storage semantics aimed at performance‑sensitive workloads, including native appends, vectored reads, and unlimited readers while writing.

Google Cloud highlighted concrete AI and analytics gains observed with Rapid Bucket: reductions in blocked GPU time of about 50%, up to 2.5× faster data loading in multi‑modal training, checkpoint restores up to 5× faster, and checkpoint writes 3.2× faster versus traditional object storage. The company positions the tier for data preparation, model training and checkpointing, batch and streaming analytics, and for optimizing distributed database architectures that need low‑latency, high‑throughput reads.

Rapid Cache, which Google first introduced at Cloud Next ’25 under the name Anywhere Cache, remains part of the Rapid family to accelerate bandwidth for AI/ML workflows. It provides on‑demand read acceleration and the ability to colocate compute with data stored in existing Cloud Storage buckets, enabling accelerated training and data preparation without requiring full dataset migration into a zonal Rapid Bucket.

Cloud Storage Rapid is positioned to address a persistent tradeoff between niche zonal storage for performance and global object stores for scale and reliability. As organizations scale training and inference across thousands of GPUs and TPUs, storage I/O has become a critical bottleneck; the Rapid family aims to minimize I/O stalls by co‑locating compute and high‑performance zonal storage while preserving Cloud Storage’s scale and simplicity.