Storage and networking highlights include 10 TB/s bandwidth with Managed Lustre and Rapid Buckets delivering up to 20 million object storage operations per second for faster checkpointing and recovery.

These storage capabilities improve checkpointing and recovery for large training runs.

Google Cloud scales training to 130,000 nodes with GKE, while GKE Agent Sandbox handles up to 300 sandboxes per second per cluster to reduce idle compute, and Cross-Cloud Network/Cloud WAN connects workloads across environments via Google's private fiber backbone spanning millions of kilometers and over 200 countries.

The platform emphasizes a flexible AI workload environment across cloud, edge, and on-premises through Cluster Director and GKE’s scalability.

Google introduced two TPU generations: TPU 8t for training, capable of connecting up to 9,600 chips per superpod, and TPU 8i for inference with 288 GB high-bandwidth memory and 384 MB on-chip SRAM.

Google backs open-source orchestration and inference tools, including llm-d, vLLM, and TorchTPU to ease migrations for PyTorch developers.

Gartner recognizes Google's AI Hypercomputer as a single-stack solution for training and inference, designed to boost utilization and cut costs as AI infrastructure spending grows.

The recognition comes amid a competitive landscape where cloud providers and chipmakers pursue a balanced mix of silicon, networking, storage, and orchestration software aimed at cost efficiency rather than raw compute alone.

Google continues collaboration with Nvidia to offer GPU-based systems through Google Cloud and plans A5X instances based on Nvidia’s Vera Rubin platform when available.

Google’s Virgo Network can connect over a million TPUs across data center sites (or up to 960,000 GPUs) in a training cluster without performance degradation, while the GKE Inference Gateway can raise model serving throughput by as much as 40% and cut costs by up to 30% through disaggregated serving and llm-d routing.

In addition, the Virgo Network supports extensive TPU and GPU scaling across sites without performance loss.

Google has been named a Leader in Gartner’s inaugural Magic Quadrant for AI Infrastructure, highlighted for its integrated AI stack, custom silicon, and execution capabilities across the Cloud.