Google Cloud’s AI Infrastructure Gets a July Upgrade

Google Cloud has been busy. This month’s AI infrastructure and orchestration updates are like a fresh current in the swamp—plenty of new life for developers to swim in. From managed Lustre to GKE scaling, here’s what you need to know.

July 2026: The Big Leaps

Managed Lustre Goes GA

Google Cloud Managed Lustre is now generally available, offering four performance tiers from 125 MB/s to 1000 MB/s per TiB, scaling up to 8 PB. Powered by DDN’s EXAScaler, it’s a solid choice for high-performance storage.

C4N VMs: Network and Storage Optimized

The new C4N VMs are GA, built to eliminate data-transfer bottlenecks. With 5th Gen Intel Xeon and Titanium offloading, they deliver 400 Gbps network bandwidth and up to 25 GiB/s block storage throughput. Perfect for data-hungry AI workloads.

GKE Dataplane V2: 15K Nodes

Standard GKE clusters can now scale to 15,000 nodes with full Network Policy enforcement. That’s a lot of pods to keep in line, but GKE’s got it covered.

Co-operative Time-Slicing in llm-d

Reinforcement learning workloads get a boost with co-operative time-slicing, increasing accelerator duty cycles from ~40% to 70% without hurting convergence. More juice from the same hardware—what’s not to love?

k8s-aibom: AI Supply Chain Security

Google open-sourced k8s-aibom, a Kubernetes controller that detects AI runtimes and generates ML-BOMs. Keep your AI supply chain clean and your shadow AI in check.

Guides and How-Tos

• Day 0 support for Kimi K3: Moonshot AI’s 2.8T-parameter model is ready to deploy on Google Cloud the day weights dropped. Get the step-by-step guide.
• GKE managed DRANET: Now supports both GPUs and TPUs, with a hands-on lab for Autopilot clusters.
• Ray on TPUs: Learn to run Ray on TPUs instead of GPUs in a two-part series.
• TPU microbenchmarks: Evaluate TPU performance with a new suite to spot bottlenecks.
• Scale agents without breaking the bank: GKE Agent Sandbox and Pod snapshots help pack more agents onto your existing compute.
• Mistral 3 optimization: Google engineers share how they boosted inference throughput by 48% on Ironwood TPUs.

Research and Reports

Google was named a Leader in the inaugural Gartner Magic Quadrant for AI Infrastructure, and their State of AI Infrastructure report reveals a widening gap between ambition and reality—83% of orgs need upgrades for agentic AI. Time to start planning.

June 2026: Security and Telemetry

Confidential Computing is now available on G4 VMs with NVIDIA RTX PRO 6000 GPUs, protecting data in use. The new TPU Developer Hub and OpenTelemetry-based telemetry agent give you better visibility into TPU performance.

May 2026: Agent Sandbox and More

GKE Agent Sandbox is GA, Agent Substrate is open-sourced, and Cloud Storage Rapid offers high-performance storage for AI. Plus, a deep dive into network infrastructure for AI workloads.

For more on Google Cloud’s platform, check out our Google Cloud review. And if you’re comparing AI models, our pricing comparison might help.

Original announcement published on Google Cloud.