Google Cloud has announced a major enhancement to Filestore, its first-party NFS file service: the storage backend now runs directly on Colossus, Google’s foundational distributed storage system. This places Filestore on the same infrastructure that powers YouTube, Gmail, and Gemini, replacing the previous VM-based architecture with a more scalable and operationally efficient platform.

The key architectural change is the decoupling of storage capacity and performance. Filestore customers can now provision IOPS independently, so they can tune performance to workload demands without over-provisioning capacity. This is particularly useful for containerized environments: through the Filestore CSI driver, the service delivers persistent high-performance storage for GKE workloads, and Filestore multishares for GKE lets teams segment a single instance into smaller shares starting at just 10 GiB. That allows AI teams to carve large high-performance instances into project-specific shares supporting thousands of concurrent containers without raising total cost of ownership.

The Colossus-backed design also enables a new use case: high-concurrency workspaces for AI agent swarms. Multiple autonomous agents often need to read and write a common dataset while maintaining shared state and context. Filestore file shares act as these common workspaces, and the NFS protocol provides consistent file system access across the swarm. NFS file locking ensures strict data consistency, and the article states the service can support millions of agents collaborating on shared data.

Operationally, the distributed storage layer brings independent scaling of IOPS via Custom Performance settings, faster failure recovery, and zero-downtime capacity changes compared to traditional monolithic architectures. Security is handled through integration with Google Cloud IAM, NFS User IDs and Group IDs, and IP Access Control Lists. The post positions this update as a foundation for future enhancements aligned with AI-era and agentic workflows.