Google Cloud has moved its Managed Lustre storage service and its C4N network- and storage-optimised virtual machines into general availability as part of a monthly update to its AI infrastructure portfolio. Managed Lustre, which uses DDN's EXAScaler technology, is offered in four performance tiers ranging from 125 MB/s to 1000 MB/s per TiB of capacity and can scale to 8 PB of storage. C4N, Google's first virtual machine series designed for network- and block storage-heavy workloads, uses 5th Gen Intel Xeon Scalable processors and Google's Titanium hardware, with 400 Gbps network bandwidth, 95 million packets per second and up to 25 GiB/s of block storage throughput when used with Hyperdisk Extreme. Google also raised the upper size limit for standard Google Kubernetes Engine clusters using Dataplane V2 with network policies enabled to 15,000 nodes. A co-operative time-slicing feature in llm-d allows reinforcement learning workloads to interleave jobs on shared systems, raising accelerator duty cycles from about 40% to 70%. Google open-sourced k8s-aibom, a Kubernetes controller that creates CycloneDX machine learning bills of materials. The update also highlighted support materials for Moonshot AI's Kimi K3, Ray on TPUs, and a blueprint for Mistral 3 Large on Ironwood, where engineering changes produced a 1.5x performance gain and increased throughput by