This month's news on AI infrastructure and orchestration

Google Cloud unveiled its July 2026 infrastructure and orchestration updates for AI and agent workloads. Key highlights include high-performance storage and network-optimized VMs, large-scale GKE clusters, improved reinforcement learning resource utilization, and AI supply chain security tools.

• Managed Lustre has reached general availability. It offers 4 performance tiers of 125, 250, 500, and 1000MB/s per capacity, and can scale up to 8PB. It is built on DDN's EXAScaler technology.
• The C4N network- and storage-optimized VM has also reached GA. It uses 5th Gen Intel Xeon Scalable processors and Titanium offload hardware, delivering up to 400Gbps network bandwidth, 95 million MPPS, and up to 25GiB/s block storage throughput when paired with Hyperdisk Extreme.
• GKE Dataplane V2 can now scale standard GKE clusters up to 15,000 nodes while retaining Network Policy. This feature targets large-scale enterprise and AI/ML infrastructure.

llm-d cooperative time-slicing has also been added for reinforcement learning workloads. It interleaves multiple independent RL jobs on shared hardware, raising accelerator utilization from about 40% to up to 70% without affecting model convergence or accuracy.

On the AI supply chain security front, the open-source Kubernetes controller k8s-aibom was released. This tool automatically detects AI runtimes such as vLLM and Triton in clusters and generates CycloneDX standard-based ML-BOMs, helping teams continuously track running AI components.