What’s new in AI infrastructure and orchestration in August
Google Cloud announced AI infrastructure updates including Filestore backend improvements, gVisor in Ray clusters, new Cloud Run instances, and Managed Lustre GA with four performance tiers.
Filestore, Google Cloud’s NFS file service, now uses a Colossus-based backend to provision IOPS independently from storage capacity and integrates with GKE, improving performance for agentic swarms that share datasets. The update addresses latency and scalability challenges in AI workflows requiring concurrent read/write operations across large agent groups.
gVisor sandboxes are now available in distributed Ray clusters on GKE through an experimental library developed with Anyscale, offering stronger isolation than standard containers with faster startup and lower memory overhead. This feature targets environments where multi-tenant AI workloads require secure, lightweight execution environments without compromising performance.
Cloud Run now offers dedicated singleton compute instances that remain active during idle periods, costing $5.70 for 30 days with 1 vCPU and 1 GiB of memory. The update supports continuous AI agent operations while reducing costs compared to traditional persistent compute models, benefiting users with sporadic workloads.
Google Cloud Managed Lustre is now generally available in four performance tiers, delivering throughput from 125 MB/s to 1000 MB/s per TiB and scaling up to 8 PB. Powered by DDN’s EXAScaler, the service targets high-performance computing and AI workloads requiring low-latency, high-throughput storage solutions.