At the Google Cloud Next conference, Google Cloud announced the Rapid series of Cloud Storage, specifically designed to accelerate AI workloads. The series includes the now generally available Rapid Bucket and the preview-phase Rapid Cache, both built on Google's proprietary Colossus file system.
Rapid Bucket: Breakthrough in Throughput and Latency
Rapid Bucket delivers over 15 TB/s of aggregated bandwidth, supports up to 20 million requests per second, and achieves sub-millisecond latency. This enables a 50% reduction in GPU blocking time and a 2.5x speedup in data loading, effectively alleviating I/O bottlenecks in large-scale AI training and real-time inference. Google Cloud emphasizes that Rapid Bucket is ideal for distributed training, high-performance computing, and latency-sensitive inference pipelines.
Rapid Cache: Accelerated Reads Without Code Changes
Targeting read-heavy workloads, Rapid Cache provides 2.5 TB/s of aggregate read throughput without requiring any application code modifications. This caching layer offers near-memory access speeds to existing storage setups, benefiting frequent reads of model checkpoints, datasets, and embedding vectors in AI pipelines.
Broader Storage Stack Upgrades and Ecosystem Expansion
Google Cloud also introduced other enhancements: automatic metadata annotation for faster data classification and retrieval, and Model Context Protocol (MCP) support enabling AI agents to directly connect with storage services. Additionally, Google Cloud NetApp Volumes and Filestore for GKE extend enterprise NAS capabilities for containerized AI workloads, providing higher-performance file storage options.
The Rapid series represents Google Cloud's latest move to strengthen its AI infrastructure offerings, differentiating through software-defined storage optimizations that deliver cost-effective data processing foundations for AI developers and enterprises.