Google Cloud AI Infrastructure and Orchestration Updates - July 2026
Google Cloud is enhancing its AI infrastructure and orchestration capabilities with several new product updates and features, including GA availability for Managed Lustre, C4N VMs, and GKE Dataplane V2 scaling. These updates aim to support the growing demands of AI and agentic workloads, offering improved performance, scalability, and security for developers and enterprises. The announcements include new security tools, practitioner guides, and research reports, all designed to help users leverage Google Cloud's AI capabilities more effectively and efficiently. This monthly update covers a range of AI-related services and tools designed for rapid innovation and cost optimization.
- →Google Cloud Managed Lustre now GA with four performance tiers
- →C4N network and storage-optimized VMs are GA
- →GKE Dataplane V2 supports up to 15K nodes with Network Policies (GA)
- →Co-operative time-slicing for reinforcement learning workloads
- →Open-sourced k8s-aibom for AI supply chain security on GKE
Features (5) ›
- Google Cloud Managed Lustre now GA with four performance tiers
Google Cloud Managed Lustre is now generally available in four distinct performance tiers, offering throughput from 125 MB/s to 1000 MB/s per TiB of capacity and scaling up to 8 PB. This solution combines DDN's EXAScaler with Google Cloud's infrastructure expertise.
- C4N network and storage-optimized VMs are GA
The new C4N VM series, optimized for network and block storage, is now generally available. Powered by 5th Gen Intel Xeon Scalable processors and Google's Titanium offloading hardware, these VMs provide 400 Gbps network bandwidth and up to 25 GiB/s block storage throughput.
- GKE Dataplane V2 supports up to 15K nodes with Network Policies (GA)
GKE Dataplane V2 now supports scaling standard GKE clusters up to 15,000 nodes while maintaining full Network Policy enforcement, addressing the large-scale infrastructure needs of enterprise and AI/ML customers.
- Co-operative time-slicing for reinforcement learning workloads
A new feature enables interleaving independent reinforcement learning jobs on shared physical hardware, potentially increasing aggregate accelerator duty cycles from approximately 40% to 70% without affecting model convergence or accuracy.
- Open-sourced k8s-aibom for AI supply chain security on GKE
k8s-aibom is a new open-source, lightweight Kubernetes controller that automatically detects AI runtimes in container clusters and generates CycloneDX Machine Learning Bill of Materials (ML-BOMs), enhancing AI supply chain security.
Notes (1) ›
- Guides available for deploying AI models and optimizing TPU usage
New practitioner guides cover topics such as Day 0 support for Moonshot AI's Kimi K3 model, running Ray on TPUs, evaluating TPU performance with microbenchmarks, and scaling agents using GKE features like Agent Sandbox.
https://cloud.google.com/blog/topics/ai-infrastructure/whats-new-in-ai-infrastructure-this-month/
Related releases
- Data Studio Conversational Analytics GA, adds BigQuery integration Google Cloud release notes ·
- Cortex Framework v7 GA with modular architecture, Dataform, and AI features Google Cloud release notes ·
- BigQuery: Audit UI query result downloads Google Cloud release notes ·
- BigQuery Data Transfer Service adds incremental Klaviyo transfers Google Cloud release notes ·
- Google Cloud Announces Borderless Lakehouse for Cross-Cloud Data Access Google Cloud Blog ·
- Cloud SDK 578.0.0: Database Migration default change, AlloyDB backup DR GA, BigQuery improvements Google Cloud release notes ·